CN117474140A - Low-voltage heavy overload classification prediction method based on convolutional neural network - Google Patents
Low-voltage heavy overload classification prediction method based on convolutional neural network Download PDFInfo
- Publication number
- CN117474140A CN117474140A CN202210883892.9A CN202210883892A CN117474140A CN 117474140 A CN117474140 A CN 117474140A CN 202210883892 A CN202210883892 A CN 202210883892A CN 117474140 A CN117474140 A CN 117474140A
- Authority
- CN
- China
- Prior art keywords
- prediction
- neural network
- load
- algorithm
- heavy overload
- 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
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 76
- 238000000034 method Methods 0.000 title claims abstract description 49
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 70
- 238000009826 distribution Methods 0.000 claims abstract description 25
- 238000011156 evaluation Methods 0.000 claims abstract description 25
- 230000002159 abnormal effect Effects 0.000 claims abstract description 6
- 238000012545 processing Methods 0.000 claims description 22
- 239000013598 vector Substances 0.000 claims description 14
- 238000013528 artificial neural network Methods 0.000 claims description 10
- 230000005611 electricity Effects 0.000 claims description 10
- 230000006870 function Effects 0.000 claims description 10
- 230000000694 effects Effects 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 4
- 230000009467 reduction Effects 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 244000309464 bull Species 0.000 claims description 2
- 238000011835 investigation Methods 0.000 claims description 2
- 238000012886 linear function Methods 0.000 claims description 2
- 230000007246 mechanism Effects 0.000 claims description 2
- 238000007781 pre-processing Methods 0.000 claims description 2
- 230000007787 long-term memory Effects 0.000 claims 1
- 230000001537 neural effect Effects 0.000 claims 1
- 238000013519 translation Methods 0.000 claims 1
- 238000005516 engineering process Methods 0.000 abstract description 3
- 238000011160 research Methods 0.000 abstract description 3
- 238000013178 mathematical model Methods 0.000 abstract description 2
- 238000012549 training Methods 0.000 description 14
- 239000011159 matrix material Substances 0.000 description 9
- 230000008859 change Effects 0.000 description 7
- 238000011176 pooling Methods 0.000 description 7
- 230000008901 benefit Effects 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 238000012986 modification Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 238000013461 design Methods 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 230000032683 aging Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012552 review Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000006467 substitution reaction 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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Biophysics (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Operations Research (AREA)
- Primary Health Care (AREA)
- Life Sciences & Earth Sciences (AREA)
- Quality & Reliability (AREA)
- Biomedical Technology (AREA)
- Entrepreneurship & Innovation (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a low-voltage heavy overload classification prediction method based on a convolutional neural network, which comprises the steps of calculating characteristic variables of a prediction model, optimizing heavy overload classification prediction of a low-voltage station area, changing the characteristic parameters, repeating the steps, establishing a plurality of different prediction models, evaluating the established prediction models to obtain evaluation parameters, comparing the evaluation parameters of each prediction model, and determining an optimal prediction model. The invention provides a research method of a 10kv low-voltage distribution transformer heavy overload classification prediction model based on dynamic evaluation and a convolution network, which takes heavy overload classification load prediction as a real target, considers weather factors and holiday abnormal data, establishes a corresponding mathematical model for heavy overload and time, adopts a model established by introducing a 3-layer technology algorithm in a transformer-CNN algorithm to solve in order to improve the accuracy of the heavy overload classification prediction model and complete tasks.
Description
Technical Field
The invention relates to the technical field of information automation, in particular to a low-voltage heavy overload classification prediction method based on a convolutional neural network.
Background
Since the 21 st century, people have continuously improved their quality of life and rapidly increased power demands, increased production and power loads, and increased power supply business to original distribution and transformer load equipment, bringing great potential hazards to the operational safety of the power transmission network. In the peak period of power consumption, particularly in summer and winter, the transformer stations in the low-voltage area are excessively stacked, the capacity of the transformer is insufficient, abnormal power supply is performed, the reliability of the power supply is affected, and economic loss is caused. Moreover, long-term overload of the equipment can accelerate ageing of the components, reduce the service life of the machine, bring the hidden danger and running risk of the power transmission network, and reduce the economic benefit of the enterprise. With the continuous increase of the power load, the reproduction of the low-voltage transformer area is frequent, and direct economic loss and huge hidden trouble are caused for the safe operation of the power grid. The traditional prediction can not achieve the effect of accuracy in time, and a large amount of monitoring time cost and technical overhaul labor cost are required. The condition of heavy overload of the low-voltage transformer area is considered to be the condition of line damage caused by equipment ageing problems, weather conditions, environmental factors and the like.
The low-voltage distribution transformer is more susceptible to the heavy overload condition caused by low temperature in winter during the peak-meeting winter, and the prediction requirement is larger. In addition, as the informatization construction of the low-voltage distribution network is carried out earlier, the method can provide better data support for the analysis and prediction model.
Disclosure of Invention
This section is intended to outline some aspects of embodiments of the invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section as well as in the description summary and in the title of the application, to avoid obscuring the purpose of this section, the description summary and the title of the invention, which should not be used to limit the scope of the invention.
The present invention has been made in view of the above-mentioned or existing problems occurring in the prior art.
It is therefore an object of the present invention to provide a low-voltage heavy overload classification prediction method based on convolutional neural networks.
In order to solve the technical problems, the invention provides the following technical scheme: a low-voltage heavy overload classification prediction method based on convolutional neural network comprises,
collecting characteristic parameters of related variables required by modeling;
calculating characteristic variables of the prediction model according to the characteristic parameters;
establishing a prediction model according to the characteristic variables;
changing characteristic parameters, repeating the steps, and establishing a plurality of different prediction models;
based on a plurality of prediction models, carrying out multiple predictions on the distribution transformer load by adopting an isodata clustering algorithm;
based on the prediction result, evaluating the established multiple prediction models to obtain evaluation parameters;
comparing the evaluation parameters of each prediction model, and determining the optimal prediction model by establishing a long short-Term Memory (LSTM) artificial neural network regression network;
and utilizing the optimal prediction model to conduct classified prediction on the low-voltage heavy overload of the convolutional neural network.
As a preferable scheme of the low-voltage heavy overload classification prediction method based on the convolutional neural network, the invention comprises the following steps: the characteristic parameters comprise load characteristic parameters, user characteristic parameters and weather characteristic parameters, wherein the load characteristic parameters are used for calculating load characteristic variables, the user characteristic parameters are used for calculating user characteristic variables, and the weather characteristic parameters are used for calculating weather characteristic variables.
As a preferable scheme of the low-voltage heavy overload classification prediction method based on the convolutional neural network, the invention comprises the following steps: calculating load characteristic variables of the mean value, the peak value, the increase and decrease and the fluctuation of the load in the load characteristic parameters by utilizing a data processing rule; the resident characteristic parameters are resident electricity utilization ratio, resident electricity utilization increase index, resident contract capacity and ratio characteristic quantity, and the resident characteristic parameters are calculated by utilizing a data processing rule to obtain resident characteristic variables; the weather characteristic parameters are weather temperature and temperature sensing, and the weather characteristic variables are calculated by using a data processing rule, wherein the data processing rule is an existing rule for mathematically processing data, so as to remove unsuitable abnormal data.
As a preferable scheme of the low-voltage heavy overload classification prediction method based on the convolutional neural network, the invention comprises the following steps: the prediction model established according to the characteristic parameters comprises a transducer-CNN algorithm model which is responsible for characteristic extraction of meteorological factors and historical load data, wherein a transducer structure consists of an encoder algorithm and a decoder algorithm.
As a preferable scheme of the low-voltage heavy overload classification prediction method based on the convolutional neural network, the invention comprises the following steps: the transducer-CNN algorithm model includes,
a layer 3 intent algorithm network was introduced that utilized a modified convolutional neural network algorithm to achieve load-rate prediction by creating an LSTM regression network for prediction of the neural network algorithm after clustering.
7. As a preferable scheme of the low-voltage heavy overload classification prediction method based on the convolutional neural network, the invention comprises the following steps: the prediction of the distribution transformer load comprises the following steps of carrying out normalization processing on a predicted load curve according to various factors influencing heavy overload of the distribution transformer to obtain:
Y i ＝[y i1 ，y i2 ...，y in ]
and (3) obtaining local trend characteristic data:
Y i '＝[y i1 '，y i2 '，…，y i(n-1) ']
wherein Y is i Representing the load data after the dimension reduction.
9. As a preferable scheme of the low-voltage heavy overload classification prediction method based on the convolutional neural network, the invention comprises the following steps: the encoder algorithm consists of 6 identical layers, wherein each two sub-layers form a layer, and the output expression of the obtained sub-layers is as follows:
sub_layer_output＝LayerNorm(x+(SubLayer(x)))
wherein, sub-layer is the bull attention layer, and the expression that it utilizes 3 layer attention algorithm to realize the load rate prediction is:
attention_output＝Attention(Q,K,V)
wherein Q, K, V is the input feature itself, three vectors generated according to the input feature, Q is the query vector, K is the vector under investigation, V is the content vector;
the h different linear variables are projected on the sub-layer, and the obtained result is:
MultiHead(Q,K,V)＝Concat(head 1 ,...,head h )W o
head i ＝Attention(QW i Q ,KW i K ,VW i V )
wherein the MultiHead-Attention model can capture more abundant characteristic information, W i Q 、W i K And W is i V Representing three weight matrices corresponding to q, k, v;
q, K and V are the same in self-attrition mechanism, wherein rated dot-product is adopted for attrition calculation, and the method comprises the following steps:
wherein, softmax represents values between 0 and 1;
before adding the result of the encoding algorithm, the decoding algorithm and the data preprocessing to the corresponding position, the sum of the encoding data and the ebedding data needs to be calculated, wherein the following formula is obtained by using sine and cosine functions to calculate respectively:
wherein PE is pos Is a linear function representing PE at any position (pos+k) 。
As a preferable scheme of the low-voltage heavy overload classification prediction method based on the convolutional neural network, the invention comprises the following steps: the establishment of the prediction model is based on an isodata clustering algorithm to establish the distribution transformer load.
As a preferable scheme of the low-voltage heavy overload classification prediction method based on the convolutional neural network, the invention comprises the following steps: the step of building a predictive model using an isodata clustering algorithm includes,
classifying the load variables into different variables by using an isodata clustering algorithm;
adding user variables, and screening out different types of variables by adopting an isodata clustering algorithm;
adding weather variables, and classifying the final data volume by using an isodata clustering algorithm.
As a preferable scheme of the low-voltage heavy overload classification prediction method based on the convolutional neural network, the invention comprises the following steps: the evaluating the plurality of predictive models includes,
introducing an evaluation index ACC, which comprises an accurate judgment rate C, an excessive false judgment rate EH and an excessive light false judgment rate EL;
the numerical calculation formula of the evaluation index ACC for the early warning effect is as follows,
ACC＝(TP+TN)/(FP+FN+FP+FN)
in the formula, TP refers to the number of correct results in positive case prediction, FP refers to the number of incorrect results in negative case prediction, TN: the opposite number of results in negative case prediction, FN: the opposite amount is predicted in the positive case.
The invention has the beneficial effects that: the invention provides a research method of a 10kv low-voltage distribution transformer heavy overload classification prediction model based on dynamic evaluation and a convolution network, which takes heavy overload classification load prediction as a real target, considers weather factors and holiday abnormal data, establishes a corresponding mathematical model for heavy overload and time, adopts a model established by introducing a 3-layer technology algorithm in a transformer-CNN algorithm to solve in order to improve the accuracy of the heavy overload classification prediction model and complete tasks.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art. Wherein:
fig. 1 is a flowchart of an implementation of a low-voltage heavy overload classification prediction method based on a convolutional neural network.
Fig. 2 is an isodata algorithm flow of a low-voltage heavy overload classification prediction method based on a convolutional neural network.
FIG. 3 is a flow chart of an improved convolutional neural network based on a convolutional neural network low-voltage heavy overload classification prediction method.
FIG. 4 is a flow of the intent normalization process of the low-voltage heavy overload classification prediction method based on convolutional neural network.
Fig. 5 is a flowchart of the normalization processing of the matching degree α0 by softmax of the low-voltage heavy overload classification prediction method based on the convolutional neural network.
FIG. 6 is a feature processing of the low-voltage heavy overload classification prediction method based on convolutional neural network.
Fig. 7 is a load prediction result based on LSTM of a low-voltage heavy overload classification prediction method based on a convolutional neural network.
FIG. 8 is a comparison of test set predictions of a low-voltage heavy overload classification prediction method based on convolutional neural networks.
FIG. 9 is a test set sample error for a convolutional neural network based low voltage heavy overload classification prediction method.
Fig. 10 is a comparison of test set load evaluation prediction results based on CNN of the low-voltage heavy overload classification prediction method based on convolutional neural network.
Fig. 11 is a test set load evaluation prediction result based on CNN of the convolutional neural network-based low-voltage heavy overload classification prediction method.
Detailed Description
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways other than those described herein, and persons skilled in the art will readily appreciate that the present invention is not limited to the specific embodiments disclosed below.
Further, reference herein to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic can be included in at least one implementation of the invention. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments.
Example 1
Referring to fig. 1 to 11, a first embodiment of the present invention provides a low-voltage heavy overload classification prediction method based on a convolutional neural network: comprising the steps of (a) a step of,
collecting characteristic parameters of related variables required by modeling;
calculating characteristic variables of the prediction model according to the characteristic parameters;
establishing a prediction model according to the characteristic variables;
changing characteristic parameters, repeating the steps, and establishing a plurality of different prediction models;
based on a plurality of prediction models, carrying out multiple predictions on the distribution transformer load by adopting an isodata clustering algorithm;
based on the prediction result, evaluating the established multiple prediction models to obtain evaluation parameters;
comparing the evaluation parameters of each prediction model, and determining an optimal prediction model by establishing an LSTM regression network;
and utilizing the optimal prediction model to conduct classified prediction on the low-voltage heavy overload of the convolutional neural network.
Specifically, the characteristic parameters of related variables required by modeling are collected to establish a transducer-CNN algorithm model, characteristic variables of a prediction model are conveniently calculated subsequently, then the transducer-CNN algorithm model is optimized for the heavy overload classification prediction of a mortgage area, the high-dimensional prediction characteristic is realized, the load rate prediction can be realized by utilizing a three-layer-of-technology algorithm network part, the input of the prediction model is a multi-dimensional time sequence, the prediction model is composed of a plurality of characteristic quantities, the historical load rate and meteorological data of a plurality of related time points are included, the power load data are collected according to the power consumption of a user, a corresponding power load change curve is created, the load change curve characteristics are hierarchically clustered by introducing an isodata-based clustering algorithm, the load change curve characteristics are hierarchically clustered by combining the load change power supply area, the load change of the platform area is divided into 2 major classes based on the isodata clustering algorithm, each 1 platform is selected by combining the load change power supply area load types of the platform area and the power supply area is divided into 2 major classes, the load change of the class is taken as a sample, the data of the sample is labeled by day load of day load rate of 96, the weather load rate is predicted by utilizing the historical load rate of the nerve load rate prediction model 2, the daily load rate of the weather load data is input by using the daily load point prediction model 2, and the daily load forecast data of 96 days is calculated by 96 days. And S4, taking the accurate judgment rate C, the excessive misjudgment rate EH and the excessive misjudgment rate EL of the evaluation index as the evaluation index ACC of the early warning effect. And predicting the corresponding load curve by the trained improved convolutional neural network model, and finally drawing a corresponding heavy overload classification prediction curve.
Example 2
Referring to fig. 1 to 6, a second embodiment of the present invention is based on the first embodiment: the characteristic parameters comprise load characteristic parameters, user characteristic parameters and weather characteristic parameters, wherein the load characteristic parameters are used for calculating load characteristic variables, the user characteristic parameters are used for calculating user characteristic variables, and the weather characteristic parameters are used for calculating weather characteristic variables.
Further, calculating load characteristic variables of the mean value, the peak value, the increase and decrease and the fluctuation of the load in the load characteristic parameters by utilizing a data processing rule; the resident characteristic parameters are resident electricity utilization ratio, resident electricity utilization increase index, resident contract capacity and ratio characteristic quantity, and the resident characteristic parameters are calculated by utilizing a data processing rule to obtain resident characteristic variables; the weather characteristic parameters are weather temperature and temperature sensing, and the weather characteristic variables are calculated by using a data processing rule, wherein the data processing rule is an existing rule for mathematically processing data, so as to remove unsuitable abnormal data.
Preferably, the predictive model established based on the characteristic parameters includes,
a transducer-CNN algorithm model responsible for feature extraction of meteorological factors and historical load data, wherein the transducer structure consists of an encoder algorithm and a decoder algorithm.
Preferably, the transducer-CNN algorithm model comprises,
a layer 3 intent algorithm network was introduced that utilized a modified convolutional neural network algorithm to achieve load-rate prediction by creating an LSTM regression network for prediction of the neural network algorithm after clustering.
Further, the predicting the load of the distribution transformer includes normalizing the predicted load curve according to a plurality of factors affecting heavy overload of the distribution transformer to obtain:
Y i ＝[Y i1 ,Y i2 ,....Y in ]
and (3) obtaining local trend characteristic data:
Y i '＝[Y i1 ',Y i2 ',,,Y i(n-1) ']
wherein Y is i Representing the load data after the dimension reduction.
Specifically, 3 layers of the attribute algorithm network part is introduced into a transform-CNN algorithm to realize load rate prediction, meanwhile, an evaluation index accurate judgment rate C, an overweight erroneous judgment rate EH and an overweight erroneous judgment rate EL are introduced as evaluation indexes ACC of early warning effects, the improved convolutional neural network algorithm, namely the transform-CNN, is utilized to optimize the low-voltage station heavy overload classification prediction, so that the heavy overload classification prediction with optimal effect is obtained, characteristic parameters are changed, the steps 2 and 3 are repeated, and a plurality of different prediction models are established; evaluating the established multiple prediction models to obtain evaluation parameters; and then, comparing the evaluation parameters of each prediction model, determining an optimal prediction model, determining the characteristic parameters influencing heavy overload of the distribution transformer during the windward peaked winter, and inputting the prediction model into a multidimensional time sequence according to various factors influencing heavy overload of the distribution transformer, wherein the prediction model consists of a plurality of characteristic quantities, including the historical load rate and the meteorological data of a plurality of related time points.
Example 3
Referring to fig. 2 to 11, a third embodiment of the present invention, based on the first two embodiments, further includes hierarchical clustering of the distribution transformer load curve features by using an isodata-based clustering algorithm, and building a prediction model, which is specifically described below,
classifying the load variables into different variables by using an isodata clustering algorithm;
adding user variables, and screening out different types of variables by adopting an isodata clustering algorithm;
adding weather variables, and classifying the final data volume by using an isodata clustering algorithm.
The method comprises the following specific steps:
a first step of creating a corresponding power load profile over time from the collected power load data of the station domain and using it as a next data set for prediction and training;
secondly, initializing collected data;
thirdly, inputting the preprocessed training set data image into an improved convolutional neural network until all input training sets are exhausted or an error function meets the accuracy requirement of load identification;
fourthly, clustering the preprocessed data by using an isodata algorithm;
specifically, the clustering isodata algorithm comprises the following steps;
selecting some different parameters as initial values, or modifying by itself in iterative process, distributing n pattern samples to each cluster center according to index, adding sample { x ] of input nn patterns i ,i＝1,2,3...,n}{x i I=1, 2, 3..n }, pre-selecting ncnc initial cluster centers { z } 1 ,z 2 ,...z Nc }{z 1 ,z 2 ,...z Nc Arbitrarily selecting its initial position from the sample, pre-selecting parameters such as KK = asThe number of cluster centers expected; θnθn=minimum number of samples in each cluster domain, θsθs=standard deviation of sample distance distribution in one cluster domain; θcθc=minimum distance between two cluster centers, less than this distance merge; LL = most logarithmic (cluster center that can merge); II = number of iterations;
further, calculating a distance index function of samples in various types, and dividing NN pattern samples into nearest clusters SjSj if
zj＝1Nj∑x∈Sjx，j＝1，2，…，Nczj＝1Nj∑x∈Sjx，j＝1，2，…，Nc
That is, x epsilon Sjx epsilon Sj when the distance between x and zj is the smallest;
further, according to the requirement, generating a new cluster center firstly splits and merges the cluster set generated last time, specifically, if the number of samples Sj < θn, discarding the sample subset, at this time Nc-1, correcting each cluster center, using the following formula,
zj＝1Nj∑x∈Sjx，j＝1，2，…，Nczj＝1NjΣx∈Sjx，j＝1，2，…，Nc
the total average distance of all pattern samples and their corresponding cluster centers is calculated, using the following formula,
preferably, the index parameters of various types are calculated again through iterative operation to reach the requirement directly through clustering results, and if the obtained results are converged, the iterative operation can be stopped;
preferably, the splitting, merging and iterative operation are distinguished, if the last iterative operation is carried out, θc=0 is carried out, the process goes to the last step, if the number of the cluster centers is not more than 1/2 of the set value, the process goes to the step 5.5, and the clustering existing before the splitting processing is carried out, and the splitting processing is not carried out even times; the standard deviation vector for the sample distance in each cluster is calculated, as follows,
σj＝(σ1j,σ2j,…,σnj)Tσj＝(σ1j,σ2j,…,σnj)T
wherein the components of the vector are
σij＝1Nj∑k＝1Nj(xik-zij)2
σij＝1Nj∑k＝1Nj(xik-zij)2
Where i=1, 2, …, n is the dimension of the sample feature vector, j=1, 2, …, nc is the number of clusters, and Nj is the number of samples in Sj; the maximum component in { σjmax, j=1, 2, …, nc } is calculated as { σjmax, j=1, 2, …, nc } and in a certain maximum component set { σjmx, j=1, 2, …, nc } if σjmax is present x >θs satisfies the following two conditions that the total number of samples in Nc.ltoreq.K2Nc.ltoreq.K2 and Sj exceeds twice or more of a prescribed value;
preferably, the distances of all cluster centers are calculated using the following formula
Dij＝||zi-zj||，i＝1,2,…,Nc-1，j＝i+1,…,
NcDij＝||zi-zj||，i＝1,2,…,Nc-1，j＝i+1,…,Nc
NcDij＝||zi-zj||，i＝1,2,…,Nc-1，j＝i+1,…,Nc
Comparing the values of Dij and thetac and re-incrementing the permutation according to the values of Dij < thetac, i.e
{Di1j1,Di2j2,…,DiLjL}{Di1 j1,Di2j2,…,DiLjL}
Combining two cluster centers ZikZik and ZjkZjk with the distance of DikjkDIkDkjk to obtain a new center:
Z～k＝1Nik+Njk[Nikzik+Njkzjk],k＝1,2,...,L
Zk～＝1Nik+Njk[Nikzik+Njkzjk],k＝1,2,...,L
z-k and Zk-are taken as average vectors, the regenerated center vectors are weighted by the number of samples in the cluster domain, different parameters are input when the iterative algorithm is not finished, and the first step is transferred; if the input parameters are unchanged, turning to the second step, and if the iterative algorithm is completed, the whole algorithm is completed correspondingly, and in the operation, the number of times after the iterative operation is completed is increased by 1.
Further, the research of the 10kv low-voltage distribution transformer overload classification prediction model based on the dynamic evaluation and the convolution network further comprises the following steps: the improved convolutional neural network algorithm Encoder adopting the transducer-CNN consists of N=6 identical layers, a transducer-CNN algorithm model is established, and the model is responsible for characteristic extraction of meteorological factors and historical load data, and Encoder, decoder forms a transducer structure. N=6 identical layers make up the encoder, two sub-layers make up one layer, residual connection and normalization are added to each sub-layer, the output of sub-layer is expressed as:
sub_layer_output＝LayerNorm(x+(SubLayer(x)))
ACC＝(TP+TN)/(FP+FN+TP+TN)
sub-layer：Multi-head self-attention
CNNs have locality and translational invariance, where the locality focuses on adjacent points in the feature map, and translational invariance is that the same matching rule is used for different regions, and although the generalized bias of CNNs makes the network perform better on a small amount of data, this also limits the performance on sufficient data, so there are some efforts to introduce the generalized bias of CNNs into the transducer to accelerate network convergence. The convolution check input characteristic superposition deviation value sweeps the input characteristic to perform matrix element multiplication, specifically adopts the following formula,
(i,j)∈{0,1,...L l+1 }
wherein the summation in the formula can be understood as cross-correlation (Cross-correlation), the deviation is b, and the convolution input of the first layer is Z l Indicating Z l+1 Representing the convolved output of layer l+1, Z l+1 Is of the size L l+1 The feature map is to pay attention to the length and width. Z (i, j) is taken as a pixel of the feature map, K represents the channel number of the feature map, and f convolves the kernel size parameter, s 0 Is the convolution step length parameter and p is the filling layer number, and the convolution kernel has the advantage of meeting the exchange lawLinear convolution has the disadvantage that the process of solving the parameters is complex, and the convolution is replaced by linear convolution kernel cross-correlation. In particular, when the convolution kernel is of size f=1, step s 0 =1, a fully connected network was constructed between the convolutions as follows:
the specific implementation flow of the transform-CNN algorithm is shown in fig. 3, where the dataset to be tested is input into a clustering algorithm to obtain a two-dimensional image, and convolved into a neural network to obtain the relationship between the corresponding power load and the time prediction.
And (3) carrying out segmentation point taking on the data set processed through clustering to prepare corresponding power-time curves, and training the curves through a training model by improving a convolutional neural network to obtain an electricity load distribution curve, wherein the time period is taken as an abscissa and the composite value is taken as an ordinate. And the electricity load distribution diagram can be judged again with the H value obtained by the evaluation index, so that the accuracy of the result is ensured, whether the relation between the electricity load of the low-voltage transformer area and the time accords with the corresponding heavy overload load is analyzed and tested, and whether the predicted electricity load is correct is verified.
The structural members for improving the convolutional neural network comprise a convolutional layer and a pooling layer, the different layers are required to be provided with corresponding operators, the functions of the operators are different, the purpose of setting up the convolutional operators is to represent the local structure of the nodes, and the purpose of setting up the pooling operators is to make the layering of the network clearer. Because the convolution operation is linear operation, the convolution operation only can represent linear mapping relation, and only the neural network with linear relation cannot be used, so that the corresponding activation function is added, a convolution operator, particularly a Fourier function of a signal x is defined, and the convolution operator is inversely transformed intoObtaining +.>Operators can be given based on the convolution theorem: x is g y＝U((U T x)Θ(U T y))；
In the above description g Is a convolution operator, and x, y each represent a signal of a node domain, and Θ represents multiplication of two vector elements.
Regarding the improvement of pooling operation of the convolutional neural network, the pooling operator is not critical when performing tasks such as node classification, link prediction and the like. Pooling is used in convolution primarily to represent the network hierarchy, e.g., g= (F, X), where F is the adjacency matrix, X is the feature matrix, for the set data the following formula is used,
D＝((G 1 ，y 1 )，(G 2 ，y2 ) ，...)
and also a corresponding set Y, using a mapping function
f:G→Y
Mapping to labels, and dividing the whole sub-graph into a plurality of non-overlapping sub-graphs by clustering to make full use of node characteristics and local structures in pooling operation, wherein each sub-graph becomes a new node after pooling, and a standard matrix in convolution is
Wherein I is M Is an identity matrix, C is an angle matrix, and H is an adjacency matrix.
The characteristic value of L is decomposed
L＝UΛU T ,Λ∈R N*N
Wherein U lambda U T Is a diagonal matrix formed by characteristic values of L, U is a Fourier basis, and then a convolution kernel Θ is used for carrying out convolution operation on signals on a graph to obtain the following components:
Θ* g x＝Θ(UΛU T )x
the above is to define one-dimensional data x E R N Operation of deconvolution toThis can be extended to multidimensional data, e.g. two-dimensional dataWherein c i For the characteristic dimension of the node, use +.>Represents the output after convolution with X, C 0 Feature dimensions after convolution for each node.
The essence of improving convolutional neural networks is to fourier transform on the network, i.e., let the node use the weighted sum of its own and neighboring nodes as a new feature of the current node. And the connection of the same kind of nodes is tighter, the structure is more complex, the same kind of nodes have the same phase characteristics, the downstream classification task is optimized, and finally, in order to ensure the weight value, the direction and the step length adjustment and the optimization among the layers of the neural network, the adjustment and the optimization can be carried out by a back propagation algorithm, the corresponding numerical value can be adjusted according to the deviation between the output value and the expected value, and the partial derivative of the weight bias of the loss function can be obtained by utilizing a chained derivation rule. In order to make the adjustment of the weight more accurate and stable, the weight can be adjusted again by a gradient descent algorithm.
In summary, 99.79% of the distribution transformer overload early warning information is reliably provided through the byes-LSTM prediction error of 2.79CNN classification prediction, and a time point load prediction value can be provided, data basis is provided for the daily operation maintenance work arrangement and asset management of the power grid, processed data is imported into the LSTM, 90 percent of the data is used for training models and is changed into Chinese character numbers for predicting the data, in order to obtain better fit and prevent training divergence, the training data is standardized to have zero mean value and unit variance, the value of a future time step of a predicted sequence is obtained, the response is designated as a training sequence with the value shifted by one time step, that is, the LSTM network learns the value of the predicted next time step at each time step of the input sequence, the predicted variable is the training sequence without the final time step, the LSTM regression network is created, the LSTM layer is designated to have 200 hidden units, setting parameters of a neural network, adjusting random parameters to proper values through Bayes to achieve the effect of optimizing an LSTM neural network, training the LSTM grid to generate a one-dimensional table, manually adding two columns of values of rated capacity and apparent power, importing the processed table into a CNN convolution network, randomly dividing a training set and a testing set, carrying out data normalization, keeping the same with an output layer structure, constructing the CNN convolution network, inputting characteristic data of an input layer, a relu activation layer full-connection layer classification layer, carrying out parameter setting, training a model, carrying out inverse normalization processing after predicting data to obtain a predicted data value, carrying out a fuzzy matrix, carrying out load evaluation prediction result based on the training set of the CNN, more reliably providing distribution transformer overload early warning information through byes-LSTM prediction error 2.79CNN classification prediction 99.79%, and being capable of providing a time point load predicted value, and a data basis is provided for the daily operation maintenance work arrangement and asset management of the power grid.
It is important to note that the construction and arrangement of the present application as shown in a variety of different exemplary embodiments is illustrative only. Although only a few embodiments have been described in detail in this disclosure, those skilled in the art who review this disclosure will readily appreciate that many modifications are possible (e.g., variations in sizes, dimensions, structures, shapes and proportions of the various elements, values of parameters (e.g., temperature, pressure, etc.), mounting arrangements, use of materials, colors, orientations, etc.) without materially departing from the novel teachings and advantages of the subject matter described in this application. For example, elements shown as integrally formed may be constructed of multiple parts or elements, the position of elements may be reversed or otherwise varied, and the nature or number of discrete elements or positions may be altered or varied. Accordingly, all such modifications are intended to be included within the scope of present invention. The order or sequence of any process or method steps may be varied or re-sequenced according to alternative embodiments. In the claims, any means-plus-function clause is intended to cover the structures described herein as performing the recited function and not only structural equivalents but also equivalent structures. Other substitutions, modifications, changes and omissions may be made in the design, operating conditions and arrangement of the exemplary embodiments without departing from the scope of the present inventions. Therefore, the invention is not limited to the specific embodiments, but extends to various modifications that nevertheless fall within the scope of the appended claims.
Furthermore, in order to provide a concise description of the exemplary embodiments, all features of an actual implementation may not be described (i.e., those not associated with the best mode presently contemplated for carrying out the invention, or those not associated with practicing the invention).
It should be appreciated that in the development of any such actual implementation, as in any engineering or design project, numerous implementation-specific decisions may be made. Such a development effort might be complex and time consuming, but would nevertheless be a routine undertaking of design, fabrication, and manufacture for those of ordinary skill having the benefit of this disclosure.
It should be noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention may be modified or substituted without departing from the spirit and scope of the technical solution of the present invention, which is intended to be covered in the scope of the claims of the present invention.
Claims (10)
1. A low-voltage heavy overload classification prediction method based on a convolutional neural network is characterized by comprising the following steps of: comprising the steps of (a) a step of,
collecting characteristic parameters of related variables required by modeling;
calculating characteristic variables of the prediction model according to the characteristic parameters;
establishing a prediction model according to the characteristic variables;
changing characteristic parameters, repeating the steps, and establishing a plurality of different prediction models;
based on a plurality of prediction models, carrying out multiple predictions on the distribution transformer load by adopting an isodata clustering algorithm;
based on the prediction result, evaluating the established multiple prediction models to obtain evaluation parameters;
comparing the evaluation parameters of each prediction model, and determining an optimal prediction model by establishing an LSTM long-term memory artificial neural network regression network;
and utilizing the optimal prediction model to conduct classified prediction on the low-voltage heavy overload of the convolutional neural network.
2. The convolutional neural network-based low-voltage heavy overload classification prediction method of claim 1, wherein: the characteristic parameters comprise load characteristic parameters, user characteristic parameters and weather characteristic parameters, wherein the load characteristic parameters are used for calculating load characteristic variables, the user characteristic parameters are used for calculating user characteristic variables, and the weather characteristic parameters are used for calculating weather characteristic variables.
3. The convolutional neural network-based low-voltage heavy overload classification prediction method of claim 2, wherein: calculating load characteristic variables of the mean value, the peak value, the increase and decrease and the fluctuation of the load in the load characteristic parameters by utilizing a data processing rule; the resident characteristic parameters are resident electricity utilization ratio, resident electricity utilization increase index, resident contract capacity and ratio characteristic quantity, and the resident characteristic parameters are calculated by utilizing a data processing rule to obtain resident characteristic variables; the weather characteristic parameters are weather temperature and temperature sensing, and the weather characteristic variables are calculated by using a data processing rule, wherein the data processing rule is an existing rule for mathematically processing data, so as to remove unsuitable abnormal data.
4. A low-voltage heavy overload classification prediction method based on a convolutional neural network as claimed in any one of claims 1 to 3, characterized in that: the prediction model established according to the characteristic parameters comprises a transducer-CNN neural machine translation convolutional neural network algorithm model which is responsible for characteristic extraction of meteorological factors and historical load data, wherein a transducer structure consists of an encoder algorithm and a decoder algorithm.
5. The convolutional neural network-based low-voltage heavy overload classification prediction method of claim 4, wherein: the transducer-CNN algorithm model includes,
a layer 3 intent algorithm network was introduced that utilized a modified convolutional neural network algorithm to achieve load-rate prediction by creating an LSTM regression network for prediction of the neural network algorithm after clustering.
6. The convolutional neural network-based low-voltage heavy overload classification prediction method of claim 5, wherein: the prediction of the distribution transformer load comprises the following steps of carrying out normalization processing on a predicted load curve according to various factors influencing heavy overload of the distribution transformer to obtain:
Y i ＝[Y i1 ,Y i2 ,....Y in ]
and (3) obtaining local trend characteristic data:
Y i '＝[Y i1 ',Y i2 ',,,Y i(n-1) ']
wherein Y is i Representing the load data after the dimension reduction.
7. The convolutional neural network-based low-voltage heavy overload classification prediction method of claim 6, wherein: the encoder algorithm consists of 6 identical layers, wherein each two sub-layers form a layer, and the output expression of the obtained sub-layers is as follows:
sub_layer_output＝LayerNorm(x+(SubLayer(x)))
wherein, sub-layer is the bull attention layer, and the expression that it utilizes 3 layer attention algorithm to realize the load rate prediction is:
attention_output＝Attention(Q,K,V)
wherein Q, K, V is the input feature itself, three vectors generated according to the input feature, Q is the query vector, K is the vector under investigation, V is the content vector;
the h different linear variables are projected on the sub-layer, and the obtained result is:
MultiHead(Q,K,V)＝Concat(head 1 ,...,head h )W o
head i ＝Attention(QW i Q ,KW i K ,VW i V )
wherein the MultiHead-Attention model can capture more abundant characteristic information, W i Q 、W i K And W is i V Representing three weight matrices corresponding to q, k, v;
q, K and V are the same in self-attrition mechanism, wherein rated dot-product is adopted for attrition calculation, and the method comprises the following steps:
wherein, softmax represents values between 0 and 1;
before adding the result of the encoding algorithm, the decoding algorithm and the data preprocessing to the corresponding position, the sum of the encoding data and the ebedding data needs to be calculated, wherein the following formula is obtained by using sine and cosine functions to calculate respectively:
wherein PE is pos Is a linear function representing PE at any position (pos+k) 。
8. The convolutional neural network-based low-voltage heavy overload classification prediction method of claim 7, wherein: the establishment of the prediction model is based on an isodata clustering algorithm to establish the distribution transformer load.
9. The convolutional neural network-based low-voltage heavy overload classification prediction method of claim 8, wherein: the step of building a predictive model using an isodata clustering algorithm includes,
classifying the load variables into different variables by using an isodata clustering algorithm;
adding user variables, and screening out different types of variables by adopting an isodata clustering algorithm;
adding weather variables, and classifying the final data volume by using an isodata clustering algorithm.
10. The convolutional neural network-based low-voltage heavy overload classification prediction method of claim 9, wherein: the evaluating the plurality of predictive models includes,
introducing an evaluation index ACC, which comprises an accurate judgment rate C, an excessive false judgment rate EH and an excessive light false judgment rate EL;
the numerical calculation formula of the evaluation index ACC for the early warning effect is as follows,
ACC＝(TP+TN)/(FP+FN+FP+FN)
in the formula, TP refers to the number of correct results in positive case prediction, FP refers to the number of incorrect results in negative case prediction, TN: the opposite number of results in negative case prediction, FN: the opposite number of results is predicted in the positive case.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210883892.9A CN117474140A (en) | 2022-07-26 | 2022-07-26 | Low-voltage heavy overload classification prediction method based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210883892.9A CN117474140A (en) | 2022-07-26 | 2022-07-26 | Low-voltage heavy overload classification prediction method based on convolutional neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117474140A true CN117474140A (en) | 2024-01-30 |
Family
ID=89636626
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210883892.9A Pending CN117474140A (en) | 2022-07-26 | 2022-07-26 | Low-voltage heavy overload classification prediction method based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117474140A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118095570A (en) * | 2024-04-17 | 2024-05-28 | 北京智芯微电子科技有限公司 | Intelligent load prediction method and system for transformer area, electronic equipment, medium and chip |
-
2022
- 2022-07-26 CN CN202210883892.9A patent/CN117474140A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118095570A (en) * | 2024-04-17 | 2024-05-28 | 北京智芯微电子科技有限公司 | Intelligent load prediction method and system for transformer area, electronic equipment, medium and chip |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111738512B (en) | Short-term power load prediction method based on CNN-IPSO-GRU hybrid model | |
| CN106650767B (en) | Flood forecasting method based on cluster analysis and real-time correction | |
| CN111311400A (en) | Modeling method and system of grading card model based on GBDT algorithm | |
| CN104809658B (en) | A kind of rapid analysis method of low-voltage distribution network taiwan area line loss | |
| CN111178611B (en) | Method for predicting daily electric quantity | |
| CN112735097A (en) | Regional landslide early warning method and system | |
| CN109300040A (en) | Overseas investment methods of risk assessment and system based on full media big data technology | |
| CN110717610A (en) | Wind power prediction method based on data mining | |
| CN116186548B (en) | Power load prediction model training method and power load prediction method | |
| Gupta et al. | Clustering-Classification based prediction of stock market future prediction | |
| CN113379116A (en) | Cluster and convolutional neural network-based line loss prediction method for transformer area | |
| CN114118269A (en) | Energy big data aggregation analysis method based on typical service scene | |
| CN112288157A (en) | Wind power plant power prediction method based on fuzzy clustering and deep reinforcement learning | |
| CN117474140A (en) | Low-voltage heavy overload classification prediction method based on convolutional neural network | |
| CN111027841A (en) | Low-voltage transformer area line loss calculation method based on gradient lifting decision tree | |
| CN113762591A (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM counterstudy | |
| CN113033898A (en) | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network | |
| CN113344589A (en) | Intelligent identification method for collusion behavior of power generation enterprise based on VAEGMM model | |
| CN112508363A (en) | Deep learning-based power information system state analysis method and device | |
| CN112819208A (en) | Spatial similarity geological disaster prediction method based on feature subset coupling model | |
| Zhang et al. | Load forecasting method based on improved deep learning in cloud computing environment | |
| CN116151799A (en) | BP neural network-based distribution line multi-working-condition fault rate rapid assessment method | |
| CN114238852A (en) | Operation data analysis method and device, storage medium and electronic equipment | |
| CN111461565A (en) | Power supply side power generation performance evaluation method under power regulation | |
| CN114254828B (en) | Power load prediction method based on mixed convolution feature extractor and GRU |
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 |