CN114357670A - Power distribution network power consumption data abnormity early warning method based on BLS and self-encoder - Google Patents

Power distribution network power consumption data abnormity early warning method based on BLS and self-encoder Download PDF

Info

Publication number
CN114357670A
CN114357670A CN202111421864.7A CN202111421864A CN114357670A CN 114357670 A CN114357670 A CN 114357670A CN 202111421864 A CN202111421864 A CN 202111421864A CN 114357670 A CN114357670 A CN 114357670A
Authority
CN
China
Prior art keywords
data
time
bls
self
encoder
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
Application number
CN202111421864.7A
Other languages
Chinese (zh)
Inventor
骆宇平
高如超
潘亮
陈业钊
刘皓杨
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangdong Electric Power Communication Technology Co Ltd
Original Assignee
Guangdong Electric Power Communication Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guangdong Electric Power Communication Technology Co Ltd filed Critical Guangdong Electric Power Communication Technology Co Ltd
Priority to CN202111421864.7A priority Critical patent/CN114357670A/en
Publication of CN114357670A publication Critical patent/CN114357670A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/18—Network design, e.g. design based on topological or interconnect aspects of utility systems, piping, heating ventilation air conditioning [HVAC] or cabling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • 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
    • 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
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J13/00—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network
    • H02J13/00002—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network characterised by monitoring
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2113/00—Details relating to the application field
    • G06F2113/04—Power grid distribution networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/12—Timing analysis or timing optimisation

Abstract

The invention discloses a power distribution network power consumption data abnormity early warning method based on BLS and a self-encoder, which comprises the following steps: collecting power utilization data through an information acquisition node, acquiring relevant characteristics influencing power utilization conditions, namely power utilization data EC, time information, external environment information of a region and self condition information of a line of the region, and then performing data preprocessing; predicting trends in electricity consumption data flow based on the sparse BLS model; establishing an abnormal detection model of the electricity consumption data current based on a self-encoder; and the abnormity early warning of the electricity utilization data flow is realized. The method can improve the efficiency on the basis of keeping the distribution of the original data, improve the accuracy of prediction and effectively reduce the complexity of the model. The method can effectively perform early warning on the electricity utilization abnormity, is different from the traditional analysis method based on historical data or real-time data, and is higher in timeliness and more intelligent.

Description

Power distribution network power consumption data abnormity early warning method based on BLS and self-encoder
Technical Field
The invention relates to the technical field of power consumption data abnormity early warning in a power distribution network, in particular to a power consumption data abnormity early warning method for the power distribution network based on BLS and a self-encoder.
Background
The power distribution network is close to the user side, the power distribution network directly distributes power to the user and supplies power to the user, and the safe and reliable operation level of the power distribution network can directly influence the power utilization quality of the user. With the rapid development of economy in China, the scale of a power distribution network becomes larger and larger due to the rapid increase of power consumption, the interconnection among the power grids in each area is enhanced day by day, and the operation environment is complex. The large-scale power grid interconnection improves the operation economy of the system and simultaneously enables the influence range of the faults of the local power grid on the whole system to be wider. Although modern power systems adopt a large number of advanced devices, protection and control means, the occurrence of power distribution network faults cannot be avoided due to the complexity of field environment and power utilization characteristics, which affects the experience of users and even causes huge losses to countries and power supply enterprises. If the abnormal power utilization condition of the power distribution network can be effectively predicted and early-warning is given out, operation and maintenance are carried out in advance, the power distribution network power failure accident can be prevented, the power supply reliability is improved, and the power distribution network power failure early-warning method has very important significance for reducing national economic loss and social disorder caused by faults.
The technical research and application of the big data of the power grid have been successful at home and abroad, but still are in the starting exploration stage. The big data of the power grid has the characteristics of huge data volume, numerous data types, large data increment, high speed and the like, and forms a large-scale data stream. The value of the data stream is reduced along with the lapse of time, how to rapidly learn the trend of the data stream from the data and send out early warning aiming at abnormal data in the data stream is of great significance to the operation and maintenance of equipment, and therefore a big data technology can be introduced into equipment abnormality detection.
The prior art mainly focuses on analyzing offline data, utilizes less real-time monitoring data of power grid state monitoring equipment, electric energy metering equipment and the like, lacks intelligent analysis on the current situation and characteristics implied by different power utilization scenes, and cannot perform accurate and effective early warning judgment, so that hidden dangers and leaks exist in operation and maintenance work. Therefore, the value of the power data is difficult to mine in the prior art when the power data has high value.
Disclosure of Invention
This section is for the purpose of summarizing some aspects of embodiments of the invention and to briefly introduce some preferred embodiments. In this section, as well as in the abstract and the title of the invention of this application, simplifications or omissions may be made to avoid obscuring the purpose of the section, the abstract and the title, and such simplifications or omissions are not intended to limit the scope of the invention.
The present invention has been made in view of the above-mentioned conventional problems.
Therefore, the technical problem solved by the invention is as follows: the prior art mainly focuses on analyzing offline data, utilizes less real-time monitoring data of power grid state monitoring equipment, electric energy metering equipment and the like, lacks intelligent analysis on the current situation and characteristics implied by different power utilization scenes, and cannot perform accurate and effective early warning judgment, so that hidden dangers and leaks exist in operation and maintenance work. Therefore, the value of the power data is difficult to mine in the prior art when the power data has high value.
In order to solve the technical problems, the invention provides the following technical scheme: collecting power utilization data through an information acquisition node, acquiring relevant characteristics influencing power utilization conditions, namely power utilization data EC, time information, external environment information of a region and self condition information of a line of the region, and then performing data preprocessing; constructing a sparse BLS model, training the sparse BLS model by utilizing the preprocessed data, and predicting the trend of the electricity data flow by utilizing the trained sparse BLS model; establishing a power consumption data flow abnormity detection model based on a self-encoder, and training the power consumption data flow abnormity detection model by utilizing preprocessed data; inputting the power consumption data current trend result obtained by the sparse BLS model into the power consumption data current abnormity detection model to solve the error after self-coding; and when the error exceeds a preset value, judging the error as an abnormal point, and realizing the abnormal early warning of the electricity consumption data current.
The invention relates to a preferable scheme of a power distribution network electricity consumption data abnormity early warning method based on BLS and a self-encoder, wherein the method comprises the following steps: the information acquisition node comprises an intelligent electric meter and a PMU.
The invention relates to a preferable scheme of a power distribution network electricity consumption data abnormity early warning method based on BLS and a self-encoder, wherein the method comprises the following steps: each piece of data contains 4-aspect attributes including the amount of electricity data EC, and the following features are supplemented using a sliding window: the average value of the electricity consumption data amount of the previous four days is 4ave _ EC, the average value of the electricity consumption data amount of the previous five days is 5ave _ EC, the maximum value of the electricity consumption data amount of the previous four days is 4max _ EC, and the minimum value of the electricity consumption data amount of the previous four days is 4min _ EC; time information, including: DATE, quarterly QUARTER, SEASON sea, day of week WEEKEND, whether it IS WEEKEND, where 1 represents WEEKEND and 0 represents weekday; the external environment information of the local area comprises: TEMPERATURE TEMPERATURE, WIND WIND, rainfall RAIN, snowfall SNOW; the self condition information of the line in the area comprises: the average LINE LOSS rate LINE _ LOSS of the LINE, the average CAPACITY-to-LOAD ratio CAPACITY _ LOAD, the OVERLOAD operation TIME length OVERLOAD _ TIME, and the fault occurrence frequency FAILURE _ FREQ.
The invention relates to a preferable scheme of a power distribution network electricity consumption data abnormity early warning method based on BLS and a self-encoder, wherein the method comprises the following steps: the data preprocessing process comprises the steps of detecting whether the data field is complete or not, removing data records containing missing values, and then preparing to be read by a subsequent system at any time; and carrying out normalization processing on the data, and mapping the data to a [0,1] value domain interval.
The invention relates to a preferable scheme of a power distribution network electricity consumption data abnormity early warning method based on BLS and a self-encoder, wherein the method comprises the following steps: the constructing of the sparse BLS model, training of the sparse BLS model by using the preprocessed data, and prediction of the trend of the electricity consumption data flow by using the trained sparse BLS model comprises that according to the attribute data, characteristics except the electricity consumption data EC are used as input X ═ X of the model1，…，xk，…,xK]The electricity consumption data volume EC is used as a target output Y, and a training data set is constructed; extracting features by using a Gauss-Bernoulli limited Boltzmann machine, wherein in the sparse BLS, N groups of feature mappings are shared, and the obtained hidden layer feature is ZN≡[Z1,…,Zn,…,ZN]，Zn＝[z1,…,zi,…,zDn]Denotes the n-th set of feature maps represented by DnThe characteristic composition comprises: initializing W1Mapping the offsets b, c of the layers and the input layer; mapping layer units i to 1,2, …, D for each groupnCalculating the condition distribution:
Figure BDA0003377684730000031
wherein, ciTo be offset, WkiDenotes xkAnd ziWeight of the connection between, σkIs a Gaussian input unit xkThe standard deviation referred to; for input layer unit K equal to 1,2, …, K, the conditional distribution is calculated:
Figure BDA0003377684730000032
wherein, bkIn order to be offset,
Figure BDA0003377684730000033
representing a positive-Taiwan distribution function; in the training process of adopting Gibbs sampling and comparison algorithm, inputtingThe weight and bias update process to the mapping layer is as follows:
Figure BDA0003377684730000034
Figure BDA0003377684730000035
Figure BDA0003377684730000036
wherein the content of the first and second substances,<～>dataand<～>modelexpressing the expectation related to the input data and the model distribution, updating parameters through iteration to reach the maximum iteration times, and storing the weight matrix W from the input layer to the mapping layer which is trained at present1And input offset vector b, mapping layer offset vector c.
The invention relates to a preferable scheme of a power distribution network electricity consumption data abnormity early warning method based on BLS and a self-encoder, wherein the method comprises the following steps: further comprising, constructing an enhanced node through GRU, which will obtain the characteristic ZN(t) as input to the enhancement layer at the t-th time point, the output of the enhancement layer is HM(t)≡[H1(t),…,Hm(t),…,HM(t)]I.e. there are M groups of enhanced nodes; the method comprises the following steps: randomly initializing connection weights and thresholds in M GRU neural networks, wherein the neural network is a three-layer neural network, the first layer is an input layer, the second layer is a hidden layer, the third layer is an output layer, the aim is to minimize a loss function MAE, and the formula is as follows:
Figure BDA0003377684730000041
where n represents the number of samples predicted, yjThe target predicted value is represented and,
Figure BDA0003377684730000042
representing an actual predicted value; calculating the output value of each gate in each memory module in the mth GRU network, wherein the formula is as follows:
sm(t)＝σ(V(s)ZN(t)+U(s)Hm(t-1))
rm(t)＝σ(V(r)ZN(t)+U(r)Hm(t-1))
Figure BDA0003377684730000043
Figure BDA0003377684730000044
wherein s ism(t) and rm(t) denotes an update gate and a reset gate at time t, respectively,
Figure BDA0003377684730000045
is the current input ZN(t) and previous time hidden layer output Hm(t-1) the summary, σ (-) is sigmoid function, tanh (-) is hyperbolic tangent function, V(s)，U(s)，V(r)，U(r)，V(h)，U(h)Is a parameter in the training process;
starting iteration, obtaining parameters in the enhancement layer through a BPNN gradient descent algorithm, and updating the weight value through the following formula in the training process:
Figure BDA0003377684730000046
Figure BDA0003377684730000047
in the above two formulas,
Figure BDA0003377684730000048
other rightsThe heavy gradient is solved as follows:
Figure BDA0003377684730000049
Figure BDA0003377684730000051
Figure BDA0003377684730000052
Figure BDA0003377684730000053
in the same way as above, the first and second,
Figure BDA0003377684730000054
Figure BDA0003377684730000055
after the iteration is finished, the weight matrix W of the trained GRU is stored2。
The invention relates to a preferable scheme of a power distribution network electricity consumption data abnormity early warning method based on BLS and a self-encoder, wherein the method comprises the following steps: the method also comprises the steps of connecting the mapping layer and the enhancement layer output together as input to the output layer, and obtaining the connection weight W through an elastic-net3The method comprises the following steps:
establishing a target loss function:
Figure BDA0003377684730000056
wherein [ Z ]N(t)|HM(t)]Representing the combination of the obtained features and the enhanced nodes, wherein gamma and rho are two hyper-parameters, the larger gamma is, the larger penalty is on W, and the less overfitting is easy to occur, and if rho is 0, the elastic-net degenerates to a ridge regression;
the solution of the elastic-net method is converted into a form similar to that of the lasso method, and then solved as follows:
Figure BDA0003377684730000057
Figure BDA0003377684730000058
wherein, the matrix [ ZN(t)|HM(t)]*And Y (t)*Are (T + M + N) × (M + N) and (T + M + N) × 1, respectively, assuming that
Figure BDA0003377684730000059
And
Figure BDA00033776847300000510
then the solution of elastic-net is the same as the solution of lasso:
Figure BDA00033776847300000511
solving the weight by using a Lars method to obtain an approximate solution:
Figure BDA0003377684730000061
the invention relates to a preferable scheme of a power distribution network electricity consumption data abnormity early warning method based on BLS and a self-encoder, wherein the method comprises the following steps: the establishing of the power consumption data current abnormity detection model based on the self-encoder comprises the steps of screening preprocessed power consumption data EC and dividing normal power consumption data into small time windows; determining a time basic window, given a time point t and a span d, at [ t-d, t]The data streams arriving in the time period form a time basic window WjFor the jth time basic window of the data stream, the time span is Wj,spanD; determining a time sliding windowThe sequence of successive time basic windows forms a time sliding window WsRecord WSi＝{Wi-n,Wi-n-1,…,WiIs the time sliding window after the ith time basic window arrives, wherein n represents the number of basic windows accommodated by one time sliding window, and the time span is WSiEvery time d passes, the time sliding window slides forwards by the distance of one time basic window; defining data within the time sliding window as g ═ g (g)1，g2，…，gm)TWherein g is1Representing data of a first sampling time point in a time sliding window, wherein m represents the number of sampling time points in a time basic window; and performing data reconstruction based on a self-encoder on the data in the time window, and training the self-encoder of the electricity consumption data quantity, wherein the self-encoder comprises:
constructing an auto-encoder, and establishing an objective loss function:
p＝senc(W41g+d1)
g′＝sdec(W42p+d2)
Figure BDA0003377684730000062
where p is a latent variable, sencAnd sdecActivation functions of encoder and decoder, respectively, W41、W42、d1And d2Is a parameter to be trained; initializing parameters W of an encoder and decoder41、W42、d1And d2；
And starting iteration, and updating the weight value by using a gradient descent method.
The invention relates to a preferable scheme of a power distribution network electricity consumption data abnormity early warning method based on BLS and a self-encoder, wherein the method comprises the following steps: the abnormal state judgment standard of the electricity consumption data flow comprises the steps that an electricity consumption data flow trend result obtained according to the sparse BLS model is input into the electricity consumption data flow abnormality detection model to solve an error after self coding, and the electricity consumption data flow trend result is electricity consumption data amount x' of each time period in the future; a threshold lambda is preset, when the error exceeds lambda, the abnormal point is judged, and the sensitivity of the system for detecting the abnormal value can be adjusted by adjusting the size of the lambda.
The invention has the beneficial effects that: the method can improve the efficiency on the basis of keeping the distribution of the original data, improve the accuracy of prediction and effectively reduce the complexity of the model. The method can effectively perform early warning on the electricity utilization abnormity, is different from the traditional analysis method based on historical data or real-time data, and is higher in timeliness and more intelligent.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise. Wherein:
fig. 1 is a structural diagram of a sparse BLS of a power consumption data anomaly early warning method for a power distribution network based on a BLS and a self-encoder according to an embodiment of the present invention;
fig. 2 is a basic flow diagram of a power consumption data anomaly early warning method for a power distribution network based on BLS and a self-encoder according to an embodiment of the present invention.
Detailed Description
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, specific embodiments accompanied with figures are described in detail below, and it is apparent that the described embodiments are a part of the embodiments of the present invention, not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making creative efforts based on the embodiments of the present invention, shall fall within the protection scope of the present invention.
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 than those specifically described and will be readily apparent to those of ordinary skill in the art without departing from the spirit of the present invention, and therefore the present invention is not limited to the specific embodiments disclosed below.
Furthermore, reference herein to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is 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.
The present invention will be described in detail with reference to the drawings, wherein the cross-sectional views illustrating the structure of the device are not enlarged partially in general scale for convenience of illustration, and the drawings are only exemplary and should not be construed as limiting the scope of the present invention. In addition, the three-dimensional dimensions of length, width and depth should be included in the actual fabrication.
Meanwhile, in the description of the present invention, it should be noted that the terms "upper, lower, inner and outer" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, and are only for convenience of describing the present invention and simplifying the description, but do not indicate or imply that the referred device or element must have a specific orientation, be constructed in a specific orientation and operate, and thus, cannot be construed as limiting the present invention. Furthermore, the terms first, second, or third are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
The terms "mounted, connected and connected" in the present invention are to be understood broadly, unless otherwise explicitly specified or limited, for example: can be fixedly connected, detachably connected or integrally connected; they may be mechanically, electrically, or directly connected, or indirectly connected through intervening media, or may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.
Example 1
Referring to fig. 1 to 2, an embodiment of the present invention provides a power distribution network power consumption data anomaly early warning method based on a BLS and a self-encoder, including:
s1: the electricity utilization data are collected through the information acquisition nodes, relevant characteristics influencing electricity utilization conditions, namely electricity utilization data volume EC, time information, external environment information of the area and line self condition information of the area are obtained, and then data preprocessing is carried out.
It should be noted that, the electricity consumption data is collected by information collection nodes such as an intelligent electric meter and a PMU, relevant characteristics affecting electricity consumption are obtained, and the electricity consumption data is transmitted to a buffer system composed of a plurality of servers for temporary storage through a communication network in a power grid, and each piece of data includes 4 attributes:
the electricity consumption data volume EC, and the sliding window are used to supplement the following features: the average value of the electricity consumption data amount of the previous four days is 4ave _ EC, the average value of the electricity consumption data amount of the previous five days is 5ave _ EC, the maximum value of the electricity consumption data amount of the previous four days is 4max _ EC, and the minimum value of the electricity consumption data amount of the previous four days is 4min _ EC;
time information, including: DATE, quarterly QUARTER, SEASON sea, day of week WEEKEND, whether it IS WEEKEND, where 1 represents WEEKEND and 0 represents weekday;
the external environment information of the local area comprises: TEMPERATURE TEMPERATURE, WIND WIND, rainfall RAIN, snowfall SNOW;
the self condition information of the line in the area comprises: the average LINE LOSS rate LINE _ LOSS of the LINE, the average CAPACITY-to-LOAD ratio CAPACITY _ LOAD, the OVERLOAD operation TIME length OVERLOAD _ TIME, and the fault occurrence frequency FAILURE _ FREQ.
Further, the data preprocessing process comprises:
detecting whether the data field is complete, removing data records containing missing values, and then preparing to be read by a later system at any time;
the data is normalized and mapped to a [0,1] range interval.
S2: and constructing a sparse BLS model, training the sparse BLS model by utilizing the preprocessed data, and predicting the trend of the electricity data flow by utilizing the trained sparse BLS model.
Note that, based on the attribute data obtained in S1, the input X ═ X [ X ] using the characteristics other than the electricity consumption data amount EC as the model1,…,xk,…,xK]The electricity consumption data volume EC is used as a target output Y, and a training data set is constructed;
extracting features by using a Gauss-Bernoulli limited Boltzmann machine (GBRBM), wherein in the sparse BLS, N groups of feature mappings are shared, and the obtained hidden layer features are ZN≡[Z1,…,Zn,…,ZN]，
Figure BDA0003377684730000091
Figure BDA0003377684730000092
Representing the nth set of feature maps represented by DnThe characteristic composition comprises:
A. initializing W1Mapping the offsets b, c of the layers and the input layer;
B. mapping layer units i to 1,2, …, D for each groupnCalculating the condition distribution:
Figure BDA0003377684730000093
wherein, ciTo be offset, WkiDenotes xkAnd ziWeight of the connection between, σkIs a Gaussian input unit xkThe standard deviation referred to;
C. for input layer unit K equal to 1,2, …, K, the conditional distribution is calculated:
Figure BDA0003377684730000094
wherein, bkIn order to be offset,
Figure BDA0003377684730000095
indicating positive scoreDistributing a function;
D. adopting Gibbs sampling, and in the training process of the comparison algorithm, the weight and bias input to the mapping layer are updated as follows:
Figure BDA0003377684730000096
Figure BDA0003377684730000101
Figure BDA0003377684730000102
wherein the content of the first and second substances,<～>dataand<～>modelrepresenting expectations relating to input data and model distributions;
E. updating parameters through iteration to reach the maximum iteration times, and storing the weight matrix W from the input layer to the mapping layer which is trained at present1And input offset vector b, mapping layer offset vector c.
An enhanced node is constructed through GRU, and the characteristic Z is obtainedN(t) as input to the enhancement layer at the t-th time point, the output of the enhancement layer is HM(t)≡[H1(t),…,Hm(t),…,HM(t)]I.e. there are M groups of enhanced nodes, including:
randomly initializing connection weights and thresholds in M GRU neural networks, wherein the neural network is a three-layer neural network, the first layer is an input layer, the second layer is a hidden layer, the third layer is an output layer, the aim is to minimize a loss function MAE, and the formula is as follows:
Figure BDA0003377684730000103
where n represents the number of samples predicted, yjThe target predicted value is represented and,
Figure BDA0003377684730000104
representing an actual predicted value;
calculating the output value of each gate in each memory module in the mth GRU network, wherein the formula is as follows:
sm(t)＝σ(V(s)ZN(t)+U(s)Hm(t-1))
rm(t)＝σ(V(r)ZN(t)+U(r)Hm(t-1))
Figure BDA0003377684730000105
Figure BDA0003377684730000106
wherein s ism(t) and rm(t) denotes an update gate and a reset gate at time t, respectively,
Figure BDA0003377684730000107
is the current input ZN(t) and previous time hidden layer output Hm(t-1) the summary, σ (-) is sigmoid function, tanh (-) is hyperbolic tangent function, V(s)，U(s)，V(r)，U(r)，V(h)，U(h)Is a parameter in the training process;
starting iteration, obtaining parameters in the enhancement layer through a BPNN (Back Propagation Neural network) gradient descent algorithm, and updating the weight value through the following formula in the training process:
Figure BDA0003377684730000108
Figure BDA0003377684730000111
in the above two formulas,
Figure BDA0003377684730000112
the gradients of the other weights are solved as follows:
Figure BDA0003377684730000113
Figure BDA0003377684730000114
Figure BDA0003377684730000115
Figure BDA0003377684730000116
in the same way as above, the first and second,
Figure BDA0003377684730000117
Figure BDA0003377684730000118
after the iteration is finished, the weight matrix W of the trained GRU is stored2。
The mapping layer and the enhancement layer output are used as input together, the mapping layer and the enhancement layer output are connected to the output layer, and the connection weight W is obtained through an elastic-net3The method comprises the following steps:
establishing a target loss function:
Figure BDA0003377684730000119
wherein [ Z ]N(t)|HM(t)]Representing the combination of the obtained features and the enhanced nodes, wherein gamma and rho are two hyper-parameters, the larger gamma is, the larger penalty is on W, and the less overfitting is easy to occur, and if rho is 0, the elastic-net degenerates to a ridge regression;
the solution of the elastic-net method is converted into a form similar to that of the lasso method, and then solved as follows:
Figure BDA00033776847300001110
Figure BDA00033776847300001111
wherein, the matrix [ ZN(t)|HM(t)]*And Y (t)*Are (T + M + N) × (M + N) and (T + M + N) × 1, respectively, assuming that
Figure BDA0003377684730000121
And
Figure BDA0003377684730000122
then the solution of elastic-net is the same as the solution of lasso:
Figure BDA0003377684730000123
solving the weight by using a Lars method to obtain an approximate solution:
Figure BDA0003377684730000124
s3: and establishing an electricity data flow abnormity detection model based on a self-encoder, and training the electricity data flow abnormity detection model by utilizing the preprocessed data.
It should be noted that, the preprocessed electricity consumption data EC are screened, and the normal electricity consumption data is divided into small time windows;
determining a time basic window, given a time point t and a span d, at [ t-d, t]The data streams arriving in the time period form a time basic window WjFor the jth time basic window of the data stream, the time span is Wj,span＝d；
Determining a time sliding window, the sequence of successive time basic windows forming a time sliding window WsRecord WSi＝{Wi-n,Wi-n-1,…,WiIs the time sliding window after the ith time basic window arrives, wherein n represents the number of basic windows accommodated by one time sliding window, and the time span is WSiEvery time d passes, the time sliding window slides forwards by the distance of one time basic window;
defining the data in the time sliding window as g ═ g1,g2,…,gm)TWherein g is1Representing data of a first sampling time point in a time sliding window, wherein m represents the number of sampling time points in a time basic window;
the data reconstruction based on the self-encoder is carried out on the data in the time window, and the training electricity consumption data volume self-encoder comprises:
constructing an auto-encoder, and establishing an objective loss function:
p＝senc(W41g+d1)
g′＝sdec(W42p+d2)
Figure BDA0003377684730000125
where p is a latent variable, sencAnd sdecActivation functions of encoder and decoder, respectively, W41、W42、d1And d2Is a parameter to be trained;
initializing parameters W of an encoder and decoder41、W42、d1And d2；
And starting iteration, and updating the weight value by using a gradient descent method.
S4: and inputting the power consumption data current trend result obtained by the sparse BLS model into a power consumption data current abnormity detection model to solve the error after self-coding.
S5: and when the error exceeds a preset value, judging the error as an abnormal point, and realizing the abnormal early warning of the electricity consumption data current.
S4-S5 show that the power utilization data current trend result obtained according to the sparse BLS model is input into a power utilization data current abnormity detection model to solve the error after self-encoding, and the power utilization data current trend result is the power utilization data quantity x' of each time period in the future;
a threshold lambda is preset, when the error exceeds lambda, the abnormal point is judged, and the sensitivity of the system for detecting the abnormal value can be adjusted by adjusting the size of the lambda.
The present invention is directed to the temporal behavior of the power consumption data stream, improved over the conventional BLS: the characteristics are extracted by adopting a Gauss-Bernoulli limited Boltzmann machine, so that the efficiency can be improved on the basis of keeping the distribution of original data; in addition, GRUs are used in the enhancement layer to capture the long-term dynamic characteristics of the electricity data flow, and therefore prediction accuracy is improved; the model also uses elastic-net to estimate connection weights, effectively reducing the complexity of the model. Besides, the weight value of the BLS is updated very quickly, and the method is more suitable for the environment of power consumption big data. The invention uses the anomaly detection model based on the self-encoder to detect the power utilization anomaly, and can effectively early warn the power utilization anomaly after the power utilization data flow prediction is combined. Compared with the traditional analysis method based on historical data or real-time data, the method is higher in timeliness and more intelligent.
Example 2
The embodiment is another embodiment of the present invention, which is different from the first embodiment, and provides a verification test of a power consumption data abnormality early warning method for a power distribution network based on BLS and a self-encoder.
In this embodiment, a training deepCoNN model is used, each data set is randomly divided into a training set, a verification set and a test set according to a ratio of 8:1:1, and after 500 training examples are trained in each 1000 batches, MSE (mean square error) is calculated on the verification set and the test data set; when the MSE on the data set is verified to be the lowest, a trained DeepCoNN model is obtained, the model is adopted to carry out comparison test with the network based on the BLS and the self-encoder provided by the invention, both algorithms use an open source software library calculated by TensorFlow5 numerical values to carry out test on an NVIDIA GeForce GTX TITAN X GPU, the test effects of the two algorithms are compared, and the results are shown in the following table:
table 1: the experimental results are shown in a comparison table.
Efficiency of Accuracy of Timeliness (time delay)
BLS-based and self-encoder 97.5％ 94.8％ 30ms
DeepCoNN model 91.1％ 88.6％ 42ms
The method can improve the efficiency on the basis of keeping the distribution of the original data, improve the accuracy of prediction and effectively reduce the complexity of the model. The method can effectively perform early warning on the electricity utilization abnormity, is different from the traditional analysis method based on historical data or real-time data, and is higher in timeliness and more intelligent.
It should be noted that the above-mentioned embodiments are only for illustrating the technical solutions of the present invention and not for limiting, 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 modifications or equivalent substitutions may be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, which should be covered by the claims of the present invention.

Claims (9)

1. A power distribution network electricity consumption data abnormity early warning method based on BLS and a self-encoder is characterized by comprising the following steps:
collecting power utilization data through an information acquisition node, acquiring relevant characteristics influencing power utilization conditions, and preprocessing the data;
constructing a sparse BLS model, training the sparse BLS model by utilizing the preprocessed data, and predicting the trend of the electricity data flow by utilizing the trained sparse BLS model;
establishing a power consumption data flow abnormity detection model based on a self-encoder, and training the power consumption data flow abnormity detection model by utilizing preprocessed data;
inputting the power consumption data current trend result obtained by the sparse BLS model into the power consumption data current abnormity detection model to solve the error after self-coding;
and when the error exceeds a preset value, judging the error as an abnormal point, and realizing the abnormal early warning of the electricity consumption data current.
2. The power distribution network electricity consumption data abnormity early warning method based on the BLS and the self-encoder as claimed in claim 1, wherein: the information acquisition node comprises an intelligent electric meter and a PMU.
3. The power distribution network electricity consumption data abnormity early warning method based on the BLS and the self-encoder as claimed in claim 1, wherein: each piece of data contains 4-aspect attributes including,
the electricity consumption data volume EC, and the sliding window are used to supplement the following features: the average value of the electricity consumption data amount of the previous four days is 4ave _ EC, the average value of the electricity consumption data amount of the previous five days is 5ave _ EC, the maximum value of the electricity consumption data amount of the previous four days is 4max _ EC, and the minimum value of the electricity consumption data amount of the previous four days is 4min _ EC;
time information, including: DATE, quarterly QUARTER, SEASON sea, day of week WEEKEND, whether it IS WEEKEND, where 1 represents WEEKEND and 0 represents weekday;
the external environment information of the local area comprises: TEMPERATURE TEMPERATURE, WIND WIND, rainfall RAIN, snowfall SNOW;
the self condition information of the line in the area comprises: the average LINE LOSS rate LINE _ LOSS of the LINE, the average CAPACITY-to-LOAD ratio CAPACITY _ LOAD, the OVERLOAD operation TIME length OVERLOAD _ TIME, and the fault occurrence frequency FAILURE _ FREQ.
4. The power distribution network power consumption data abnormity early warning method based on the BLS and the self-encoder as claimed in claim 1 or 3, wherein: the data pre-processing procedure comprises that,
detecting whether the data field is complete or not, removing data records containing missing values, and then preparing to be read by a later system at any time;
and carrying out normalization processing on the data, and mapping the data to a [0,1] value domain interval.
5. The power distribution network electricity consumption data abnormity early warning method based on the BLS and the self-encoder as claimed in claim 4, wherein: the construction of the sparse BLS model, the training of the sparse BLS model by utilizing the preprocessed data, the prediction of the trend of the electricity data flow by utilizing the trained sparse BLS model,
according to the attribute data, a feature other than the electricity consumption data amount EC is used as an input X ═ X of the model1,…,xk,…,xK]The electricity consumption data volume EC is used as a target output Y, and a training data set is constructed;
using a Gauss-Bernoulli limited glassThe characteristic is extracted by the Zeeman machine, in the sparse BLS, N groups of characteristic mapping are shared, and the obtained hidden layer characteristic is ZN≡[Z1,…,Zn,…,ZN]，
Figure FDA0003377684720000027
Representing the nth set of feature maps represented by DnThe characteristic composition comprises:
initializing W1Mapping the offsets b, c of the layers and the input layer;
mapping layer units i to 1,2, …, D for each groupnCalculating the condition distribution:
Figure FDA0003377684720000021
wherein, ciTo be offset, WkiDenotes xkAnd ziWeight of the connection between, σkIs a Gaussian input unit xkThe standard deviation referred to;
for input layer unit K equal to 1,2, …, K, the conditional distribution is calculated:
Figure FDA0003377684720000022
wherein, bkIn order to be offset,
Figure FDA0003377684720000026
representing a positive-Taiwan distribution function;
adopting Gibbs sampling, and in the training process of the comparison algorithm, the weight and bias input to the mapping layer are updated as follows:
Figure FDA0003377684720000023
Figure FDA0003377684720000024
Figure FDA0003377684720000025
wherein the content of the first and second substances,<～>dataand<～>modelexpressing the expectation related to the input data and the model distribution, updating parameters through iteration to reach the maximum iteration times, and storing the weight matrix W from the input layer to the mapping layer which is trained at present1And input offset vector b, mapping layer offset vector c.
6. The power distribution network electricity consumption data abnormity early warning method based on the BLS and the self-encoder as claimed in claim 4, wherein: also comprises the following steps of (1) preparing,
an enhanced node is constructed through GRU, and the characteristic Z is obtainedN(t) as input to the enhancement layer at the t-th time point, the output of the enhancement layer is HM(t)≡[H1(t),…,Hm(t),…,HM(t)]I.e. there are M groups of enhanced nodes; the method comprises the following steps:
randomly initializing connection weights and thresholds in M GRU neural networks, wherein the neural network is a three-layer neural network, the first layer is an input layer, the second layer is a hidden layer, the third layer is an output layer, the aim is to minimize a loss function MAE, and the formula is as follows:
Figure FDA0003377684720000031
where n represents the number of samples predicted, yjThe target predicted value is represented and,
Figure FDA0003377684720000032
representing an actual predicted value;
calculating the output value of each gate in each memory module in the mth GRU network, wherein the formula is as follows:
sm(t)＝σ(V(s)ZN(t)+U(s)Hm(t-1))
rm(t)＝σ(V(r)ZN(t)+U(r)Hm(t-1))
Figure FDA0003377684720000033
Figure FDA0003377684720000034
wherein s ism(t) and rm(t) denotes an update gate and a reset gate at time t, respectively,
Figure FDA0003377684720000035
is the current input ZN(t) and previous time hidden layer output Hm(t-1) the summary, σ (-) is sigmoid function, tanh (-) is hyperbolic tangent function, V(s)，U(s)，V(r)，U(r)，V(h)，U(h)Is a parameter in the training process;
starting iteration, obtaining parameters in the enhancement layer through a BPNN gradient descent algorithm, and updating the weight value through the following formula in the training process:
Figure FDA0003377684720000036
Figure FDA0003377684720000037
in the above two formulas,
Figure FDA0003377684720000038
the gradients of the other weights are solved as follows:
Figure FDA0003377684720000041
Figure FDA0003377684720000042
Figure FDA0003377684720000043
Figure FDA0003377684720000044
in the same way as above, the first and second,
Figure FDA0003377684720000045
after the iteration is finished, the weight matrix W of the trained GRU is stored2。
7. The power distribution network electricity consumption data abnormity early warning method based on the BLS and the self-encoder as claimed in claim 4, wherein: also comprises the following steps of (1) preparing,
the mapping layer and the enhancement layer output are used as input together, the mapping layer and the enhancement layer output are connected to the output layer, and the connection weight W is obtained through an elastic-net3The method comprises the following steps:
establishing a target loss function:
Figure FDA0003377684720000046
wherein [ Z ]N(t)|HM(t)]Representing the combination of the obtained features and the enhanced nodes, wherein gamma and rho are two hyper-parameters, the larger gamma is, the larger penalty is on W, and the less overfitting is easy to occur, and if rho is 0, the elastic-net degenerates to a ridge regression;
the solution of the elastic-net method is converted into a form similar to that of the lasso method, and then solved as follows:
Figure FDA0003377684720000047
Figure FDA0003377684720000048
wherein, the matrix [ ZN(t)|HM(t)]*And Y (t)*Are (T + M + N) × (M + N) and (T + M + N) × 1, respectively, assuming that
Figure FDA0003377684720000049
And
Figure FDA00033776847200000410
then the solution of elastic-net is the same as the solution of lasso:
Figure FDA0003377684720000051
solving the weight by using a Lars method to obtain an approximate solution:
Figure FDA0003377684720000052
8. the power distribution network electricity consumption data abnormity early warning method based on the BLS and the self-encoder as claimed in claim 4, wherein: the establishing of the abnormal detection model of the electricity utilization data flow based on the self-encoder comprises the following steps,
screening the preprocessed power utilization data EC, and dividing the normal power data into small time windows;
determining a time basic window, given a time point t and a span d, at [ t-d, t]The data streams arriving in the time period form a time basic window WjFor the jth time basic window of the data stream, the time span is Wj,span＝d；
Determining a time sliding window, the sequence of successive time basic windows forming a time sliding window WsRecord WSi＝{Wi-n,Wi-n-1,…,WiIs the time sliding window after the ith time basic window arrives, wherein n represents the number of basic windows accommodated by one time sliding window, and the time span is WSiEvery time d passes, the time sliding window slides forwards by the distance of one time basic window;
defining data within the time sliding window as g ═ g (g)1,g2,…,gm)TWherein g is1Representing data of a first sampling time point in a time sliding window, wherein m represents the number of sampling time points in a time basic window;
and performing data reconstruction based on a self-encoder on the data in the time window, and training the self-encoder of the electricity consumption data quantity, wherein the self-encoder comprises:
constructing an auto-encoder, and establishing an objective loss function:
p＝senc(W41g+d1)
g′＝sdec(W42p+d2)
Figure FDA0003377684720000053
where p is a latent variable, sencAnd sdecActivation functions of encoder and decoder, respectively, W41、W42、d1And d2Is a parameter to be trained;
initializing parameters W of an encoder and decoder41、W42、d1And d2；
And starting iteration, and updating the weight value by using a gradient descent method.
9. The power distribution network electricity consumption data abnormity early warning method based on the BLS and the self-encoder as claimed in any one of claims 1 and 5 to 8, wherein: the abnormal state judgment criteria of the electricity consumption data flow comprise,
inputting the power utilization data current trend result obtained by the sparse BLS model into the power utilization data current abnormity detection model to solve the error after self encoding, wherein the power utilization data current trend result is the power utilization data amount x' of each time period in the future;
a threshold lambda is preset, when the error exceeds lambda, the abnormal point is judged, and the sensitivity of the system for detecting the abnormal value can be adjusted by adjusting the size of the lambda.
CN202111421864.7A 2021-11-26 2021-11-26 Power distribution network power consumption data abnormity early warning method based on BLS and self-encoder Pending CN114357670A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111421864.7A CN114357670A (en) 2021-11-26 2021-11-26 Power distribution network power consumption data abnormity early warning method based on BLS and self-encoder

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111421864.7A CN114357670A (en) 2021-11-26 2021-11-26 Power distribution network power consumption data abnormity early warning method based on BLS and self-encoder

Publications (1)

Publication Number Publication Date
CN114357670A true CN114357670A (en) 2022-04-15

Family

ID=81096362

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111421864.7A Pending CN114357670A (en) 2021-11-26 2021-11-26 Power distribution network power consumption data abnormity early warning method based on BLS and self-encoder

Country Status (1)

Country Link
CN (1) CN114357670A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115291108A (en) * 2022-06-27 2022-11-04 东莞新能安科技有限公司 Data generation method, device, equipment and computer program product
CN116577671A (en) * 2023-07-12 2023-08-11 中国华能集团清洁能源技术研究院有限公司 Battery system abnormality detection method and device

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115291108A (en) * 2022-06-27 2022-11-04 东莞新能安科技有限公司 Data generation method, device, equipment and computer program product
CN116577671A (en) * 2023-07-12 2023-08-11 中国华能集团清洁能源技术研究院有限公司 Battery system abnormality detection method and device
CN116577671B (en) * 2023-07-12 2023-09-29 中国华能集团清洁能源技术研究院有限公司 Battery system abnormality detection method and device

Similar Documents

Publication Publication Date Title
CN109902801B (en) Flood collective forecasting method based on variational reasoning Bayesian neural network
CN112348271A (en) Short-term photovoltaic power prediction method based on VMD-IPSO-GRU
CN110705743B (en) New energy consumption electric quantity prediction method based on long-term and short-term memory neural network
CN110110912B (en) Photovoltaic power multi-model interval prediction method
CN112529282A (en) Wind power plant cluster short-term power prediction method based on space-time graph convolutional neural network
CN110648014B (en) Regional wind power prediction method and system based on space-time quantile regression
CN108388962B (en) Wind power prediction system and method
CN113554466B (en) Short-term electricity consumption prediction model construction method, prediction method and device
CN106251001A (en) A kind of based on the photovoltaic power Forecasting Methodology improving fuzzy clustering algorithm
CN114357670A (en) Power distribution network power consumption data abnormity early warning method based on BLS and self-encoder
de Guia et al. Using stacked long short term memory with principal component analysis for short term prediction of solar irradiance based on weather patterns
CN112329990A (en) User power load prediction method based on LSTM-BP neural network
CN113837499A (en) Ultra-short-term wind power prediction method and system
CN116128141B (en) Storm surge prediction method and device, storage medium and electronic equipment
Wang et al. Multi-step ahead short-term predictions of storm surge level using CNN and LSTM network
CN114492922A (en) Medium-and-long-term power generation capacity prediction method
CN115220133B (en) Rainfall prediction method, device and equipment for multiple meteorological elements and storage medium
CN112819189A (en) Wind power output prediction method based on historical predicted value
CN115329930A (en) Flood process probability forecasting method based on mixed deep learning model
CN116341613A (en) Ultra-short-term photovoltaic power prediction method based on Informar encoder and LSTM
Alharbi et al. Short-term wind speed and temperature forecasting model based on gated recurrent unit neural networks
CN117113202A (en) Power loop energy consumption detection method and equipment based on joint error stacking model
CN115877483A (en) Typhoon path forecasting method based on random forest and GRU
Chen et al. Accounting information disclosure and financial crisis beforehand warning based on the artificial neural network
CN114741952A (en) Short-term load prediction method based on long-term and short-term memory network

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