CN116742623B - Dynamic state estimation method and system based on model data dual-drive - Google Patents
Dynamic state estimation method and system based on model data dual-drive Download PDFInfo
- Publication number
- CN116742623B CN116742623B CN202310999963.6A CN202310999963A CN116742623B CN 116742623 B CN116742623 B CN 116742623B CN 202310999963 A CN202310999963 A CN 202310999963A CN 116742623 B CN116742623 B CN 116742623B
- Authority
- CN
- China
- Prior art keywords
- node
- power
- information
- abnormal
- state estimation
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 80
- 238000005259 measurement Methods 0.000 claims abstract description 127
- 230000002159 abnormal effect Effects 0.000 claims abstract description 119
- 230000009977 dual effect Effects 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims description 31
- 230000008569 process Effects 0.000 claims description 26
- 239000002245 particle Substances 0.000 claims description 25
- 238000013528 artificial neural network Methods 0.000 claims description 22
- 238000005520 cutting process Methods 0.000 claims description 22
- 238000004364 calculation method Methods 0.000 claims description 20
- 230000006870 function Effects 0.000 claims description 18
- 238000002347 injection Methods 0.000 claims description 11
- 239000007924 injection Substances 0.000 claims description 11
- 238000013507 mapping Methods 0.000 claims description 8
- 238000012546 transfer Methods 0.000 claims description 7
- 238000011144 upstream manufacturing Methods 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 6
- 210000002569 neuron Anatomy 0.000 claims description 4
- 210000004027 cell Anatomy 0.000 claims description 3
- 230000001537 neural effect Effects 0.000 claims description 3
- 238000001914 filtration Methods 0.000 description 12
- 238000004422 calculation algorithm Methods 0.000 description 6
- 239000000243 solution Substances 0.000 description 6
- 230000005540 biological transmission Effects 0.000 description 5
- 230000015654 memory Effects 0.000 description 5
- 230000005856 abnormality Effects 0.000 description 4
- 238000001514 detection method Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 3
- 238000003491 array Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 239000000835 fiber Substances 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000001902 propagating effect Effects 0.000 description 2
- 240000005528 Arctium lappa Species 0.000 description 1
- RYGMFSIKBFXOCR-UHFFFAOYSA-N Copper Chemical compound [Cu] RYGMFSIKBFXOCR-UHFFFAOYSA-N 0.000 description 1
- 238000012952 Resampling Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000001364 causal effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 229910052802 copper Inorganic materials 0.000 description 1
- 239000010949 copper Substances 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- 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
- H02J3/001—Methods to deal with contingencies, e.g. abnormalities, faults or failures
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R21/00—Arrangements for measuring electric power or power factor
- G01R21/001—Measuring real or reactive component; Measuring apparent energy
- G01R21/002—Measuring real component
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R21/00—Arrangements for measuring electric power or power factor
- G01R21/001—Measuring real or reactive component; Measuring apparent energy
- G01R21/003—Measuring reactive component
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- 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
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Power Engineering (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a dynamic state estimation method and a system based on model data dual drive, wherein the method comprises the following steps: acquiring real-time node power measurement information of the power system, and identifying current moment abnormal information, wherein the corresponding node is an abnormal point; obtaining a power predicted value at an abnormal point by adopting a data driving prediction model; obtaining a power predicted value at an abnormal point based on a model-driven prediction model; learning the predicted value to obtain a final power predicted result at the abnormal point; and taking the final power prediction result as the pseudo-measurement of the current moment, replacing the abnormal information of the current moment, and carrying out state estimation by adopting all node power measurement information of the current moment after the abnormal information is replaced, so as to realize dynamic state estimation based on model data dual-drive. The invention improves the reliability of input data and the accuracy of state estimation measurement, and can reflect the actual power system state.
Description
Technical Field
The invention belongs to the technical field of dynamic state estimation of an electric power system, and relates to a dynamic state estimation method and system based on model data dual driving.
Background
The state estimation of the power system is a precondition for maintaining the safe operation of the power system, but in the actual distribution network, due to the lack of a measuring device, random errors and other reasons, the measured data has the problems of data missing and bad data, the reliability of the state estimation is reduced, and the method for detecting the bad data by the secondary linear state estimation is available at present, but the calculation efficiency of the method is not high, and the real-time requirement of the distribution network on the detection quantity of the power system is difficult to meet, so that the abnormal detection of the real-time measured data is needed.
For the detected abnormal data, prediction replacement is needed based on historical data, the conventional prediction method comprises two types of model driving and data driving, the model driving is based on a physical model, the model driving has strong interpretation and causal logic, the generalization performance of the model is good, but as the factors mainly considered by the model driving are load factors and consumption requirements, the prediction precision of the model is low; the deep neural network based on data driving has the basic principle that the deep nonlinear neural network based on deep is learned from a large number of high-dimensional samples to realize approximation of complex functions, prediction accuracy is higher than that of model driving, but data driving has high dependence on data, so that the requirement on data quality is strict, and generalization performance is weaker than that of model driving. Therefore, a method needs to be found to be capable of fusing two driving methods, realizing data prediction with both precision and generalization performance, and providing powerful data support for subsequent state estimation with higher precision.
Meanwhile, aiming at the problem of non-Gaussian measurement noise in a power distribution network, a conventional state estimation method based on a Kalman filter frame is provided, wherein all assumed noises comprise conditional probability, joint probability and the like which are Gaussian distribution, although analytic solutions in a state estimation calculation process are convenient to obtain, all probabilities are probably not Gaussian distribution for a nonlinear non-Gaussian dynamic system, which is particularly common in the power distribution network, so that analytic solutions cannot be obtained in the state estimation, the conventional state frame based on the Kalman filter cannot be suitable for the power distribution network, and therefore, in order to solve the problem of non-Gaussian noise in power distribution network measurement data, a model needs to be found, which can solve the problem that Kalman filter is easily limited by a noise model, and has good practical application value.
Disclosure of Invention
In order to solve the defects in the prior art, the invention provides a dynamic state estimation method and a system based on model data dual drive, which can detect abnormality of real-time measurement data, realize accurate prediction of measurement by fusing a model data prediction driving method, solve the problem of non-Gaussian noise of a distribution network and improve the accuracy of state estimation of a power system.
The invention adopts the following technical scheme.
A dynamic state estimation method based on model data dual drive comprises the following steps:
s1: acquiring real-time node power measurement information of the power system, carrying out anomaly score calculation on the real-time node power measurement information by combining the historical node power measurement information, and identifying the current moment anomaly information, wherein the corresponding node is an anomaly point;
s2: obtaining a power predicted value at an abnormal point by adopting historical node power measurement information and a pre-trained data driving prediction model based on an LSTM neural network;
s3: carrying out real-time classified prediction on the load based on a model driving prediction model to obtain a power prediction value at an abnormal point;
s4: learning the predicted values obtained by the S2 and the S3 by combining the historical node power measurement information to obtain a final power predicted result at the abnormal point;
s5: and taking the final power prediction result as the pseudo-measurement of the current moment, replacing the abnormal information of the current moment, and carrying out state estimation by adopting all node power measurement information of the current moment after the abnormal information is replaced, so as to realize dynamic state estimation based on model data dual-drive.
Preferably, in S1, the node power measurement information includes node injection power and line power, wherein the node injection power includes injection active power and reactive power of the node, and the line power includes active power and reactive power of the line.
Preferably, in S1, the process of identifying the abnormality information includes:
1) Dividing the power measurement information of the historical node into four measurement sets of active power and reactive power which are injected, active power and reactive power of a line, randomly selecting M information samples from each set, taking the M information samples as root nodes of a tree, randomly generating cutting points, dividing the samples into two parts by taking the information size of the cutting points as a standard, placing the two parts on two sides of the cutting points, and repeatedly cutting in the sub-nodes until one information data is finally left or the sub-nodes reach a limited height; generating cutting points and repeating cutting operation circularly and randomly to generate K trees;
2) And calculating the average height of K trees generated by each set, further calculating the abnormal score of the sample information in the set, and indicating that the sample information is abnormal if the abnormal score exceeds a set threshold.
Preferably, for the measurement set g, itSample information anomaly scoreThe method comprises the following steps:
(3)
(2)
wherein h (g) is the average of the path length of each sample point in the tree corresponding to the measurement set g from the root node to the node where the sample point is located with respect to the sample number m;
e (h (g)) is the average of h (g) of K trees.
Average height for m samples;
，is Euler constant.
Preferably, in S2, when the data-driven prediction model based on the LSTM neural network is pre-trained, the intra-neuron information transfer of the LSTM neural network follows the following formula:
(5)
(6)
(7)
(8)
(9)
(10)
wherein,representing a forget gate, an input gate and an output gate respectively;
、、the weights of the forget gate, the input gate and the output gate are respectively represented;
、、、gate control unit bias values respectively representing states of a forgetting gate, an input gate, an output gate and a neural unit;
an input vector representing an LSTM cell;
、the states of neurons at the previous moment and the current moment are represented respectively;
representing the state of the input node at the current time t;
、the hidden layer state variables at the previous moment and the current moment are represented respectively;
and Tanh represent Sigmoid and Tanh functions, respectively.
Preferably, in S3, a model driven prediction model is adopted, an active power injection predicted value at an abnormal point is obtained according to the load type of the node where the abnormal information is located, a reactive power injection predicted value at the node where the abnormal information is located is calculated on the basis of a power factor, and a branch power information predicted value where the abnormal point is located is obtained through load flow calculation.
Preferably, the injected active power prediction value at the abnormal point is:
(12)
(11)
wherein,injecting active power into a node i where abnormal information at the moment t is located;
C represents the number of load types of the node i where the abnormal information is located;
the total node number served by the node r of the upstream substation of the node i where the abnormal information at the moment t is located;
active power of a load with the type p of the node i where the abnormal information is located at the moment t is shown;
the telemetering power of a node r of the transformer substation upstream of the node where the abnormal information is located at the moment t is shown;
a load model coefficient representing a load of type p at time t;
representing the average value of the load model coefficients;
representing the consumption demand of a load of type p at node i.
Preferably, the calculation formula of the predicted value of the injected reactive power of the node where the abnormal information is located is as follows:
wherein,injecting reactive power into a node i where abnormal information is located at the moment t;
active power of a load with the type p of the node i where the abnormal information is located at the moment t is shown;
c represents the number of load types of the node i where the abnormal information is located;
the load type at the node i ispIs a power factor of (c).
Preferably, the specific process of S4 is:
the power predicted values at the abnormal points obtained by the S2 and the S3 are adopted to replace the abnormal information in the S1 real-time node power measurement information, two large groups of training sets are formed by combining the historical node power measurement information in the S2, and the following prediction operation is carried out on each large group of training sets: dividing the training set into a plurality of subgroups for training a plurality of base learners, wherein the number of subgroups is the same as the number of base learners; training the corresponding basic learners by adopting each group of data, and outputting respective prediction results by each basic learner;
The outputs of all the base learners are trained as inputs to the meta-learner, the outputs of the meta-learner being the power predictions at the outliers.
Preferably, in S5, all node power measurement information at the current time after the replacement of the abnormality information is adoptedThe model for performing the state estimation is:
(13)
(14)
wherein,a state estimation value representing the time t;
a state transfer function representing time t;
，respectively representing process noise and measurement noise;
a nonlinear mapping function for measuring a state quantity based on the quantity of the particle filter.
The non-linear mapping logic from measurement to state quantity is:
the state posterior probability is adopted to represent a state estimation result, and the posterior probability is obtained through discrete sample points obtained through random sampling, so that the state estimation result represented by a particle weight polynomial is obtained; solving the particle weight, removing the items with the particle weight smaller than the set value, and summing the rest items to obtain a final state estimation result.
A model data dual drive based dynamic state estimation system comprising:
the information acquisition and anomaly identification module is used for acquiring real-time node power measurement information of the power system, carrying out anomaly score calculation on the real-time node power measurement information by combining the historical node power measurement information, and identifying the current moment anomaly information, wherein the corresponding node is an anomaly point;
The data driving prediction module is used for obtaining a power prediction value at an abnormal point by adopting historical node power measurement information and a pre-trained data driving prediction model based on the LSTM neural network;
the model driving prediction module is used for carrying out real-time classified prediction on the load based on the model driving prediction model to obtain a power prediction value at an abnormal point;
the integrated learning module is used for learning the predicted value obtained by the data driving prediction module and the model driving prediction module by combining the historical node power measurement information to obtain a final power predicted result at an abnormal point;
and the state estimation module is used for taking the final power prediction result as the pseudo-quantity measurement of the current moment, replacing the abnormal information of the current moment, and carrying out state estimation by adopting all node power measurement information of the current moment after the abnormal information is replaced, so as to realize the dynamic state estimation based on the model data double driving.
A terminal comprising a processor and a storage medium; the storage medium is used for storing instructions;
the processor is configured to operate in accordance with the instructions to perform the steps of the method.
A computer readable storage medium having stored thereon a computer program which when executed by a processor realizes the steps of the method.
The invention has the beneficial effects that compared with the prior art:
1) According to the method, the problem that the distribution network measurement data is easy to be abnormal is considered, the abnormal score calculation is carried out on the real-time node power measurement information by combining the historical node power measurement information so as to detect the missing data and the error data, and the reliability of the input data is improved;
2) According to the invention, the data driving prediction based on the LSTM neural network, the model driving prediction and the integrated learning are combined to realize the model data mixed prediction, the node power is predicted from two parts of data trend and actual physical significance, the data driving of the invention can further mine the characteristics contained in the data, finally realize the prediction, the generalization potential of the model is improved, the model driving of the invention builds a model with actual physical significance according to the detection information of the transformer substation, and the problem of higher requirement on the data quality of the traditional data driving is solved. The optimal pseudo quantity measurement is obtained through final integrated learning of the double-drive model, the influence of bad data on a state estimation result is reduced, and the accuracy of state estimation measurement is improved;
3) According to the invention, the condition that the traditional Kalman filtering method is limited by the noise model, the measurement data of the power distribution network under the non-Gaussian noise interference is difficult to process is considered, the particle filtering is introduced to process the noise, and the condition estimation is carried out based on the particle filtering, so that the noise distribution is not limited by the model, and the actual power system condition can be reflected more.
Drawings
FIG. 1 is a flow chart of a dynamic state estimation method based on model data dual driving of the present application;
FIG. 2 is a block diagram of a standard IEEE33 node distribution network system;
FIG. 3 is a graph of the filtering results of different nodes under the influence of Laplace noise;
fig. 4 is a graph of comparison of the results of different algorithm estimates in the presence of outlier data.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the technical solutions of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. The described embodiments of the application are only some, but not all, embodiments of the application. All other embodiments, which can be made by those skilled in the art without making any inventive effort, are within the scope of the present application.
As shown in fig. 1, embodiment 1 of the present application provides a dynamic state estimation method based on model data dual driving, which in a preferred but non-limiting embodiment of the present application comprises the following steps:
s1: acquiring real-time node power measurement information of the power system, carrying out anomaly score calculation on the real-time node power measurement information by combining the historical node power measurement information, and identifying the current moment anomaly information, wherein the corresponding node is an anomaly point;
S1.1: acquiring real-time node power measurement information of a power system;
the method comprises the steps of acquiring real-time initial information of a power system, wherein the initial information is mainly node power measurement information;
the set of node measurement information may be represented as follows:
(1)
wherein,the node power measurement information set is used for representing a set of all quantity measurement of the node i at the moment of 0-T;
is a line between the i node and the j node;
n, T the number of nodes and the measurement period;
and->Injecting power into the node, wherein the power respectively represents the active power and the reactive power of the node i at the moment t;
and->For the line power, the line +.>Active power and reactive power of (a).
S1.2: carrying out anomaly score calculation on the real-time node power measurement information by combining the historical node power measurement information, and identifying the current moment anomaly information, wherein the corresponding node is an anomaly point;
the anomaly information is mainly embodied in data missing and data anomaly.
The step can rapidly detect abnormal data in big data and is mainly divided into two stages of training and evaluation, and is specifically as follows:
(1) in offline training, historical node power measurement information is divided into、、And->For each set, randomly selecting M information samples from the four measurement sets, taking the M information samples as root nodes of a tree, randomly generating a cutting point o, dividing the samples into two parts by taking the information size as a standard, placing the two parts on two sides of the cutting point, placing a node smaller than the data of the cutting point on the left side of the cutting point, otherwise placing the node on the right side, and repeatedly cutting in the rest child nodes until one piece of information data is finally left or the child nodes reach a limited height; wherein all nodes except the root node are called child nodes; usually, the maximum value (limited height) is not set first, and is usually the average height of the tree. For the mature isolated forest anomaly detection algorithm package, the defined height can be automatically determined by a program, and therefore is not described in the present invention.
The method comprises the steps of circularly and randomly generating cutting points and the following operation, namely circularly and randomly generating the cutting points, dividing a sample into two parts by taking the information size of the cutting points as a standard, placing the two parts on two sides of the cutting points, repeatedly cutting in the child nodes until finally one piece of information data is left or the child nodes reach a limited height, generating K trees, and setting the number of the trees to be 100 according to experience.
(2) In the evaluation, the real-time node power measurement information is divided into、、And->Four measurement sets, one by oneAnd substituting the number into a set of the corresponding tree to calculate the average height, further calculating the abnormal score of the sample information in the set, and indicating that the sample information is abnormal if the abnormal score exceeds a set threshold.
Dividing real-time node power measurement information into four measurement sets of injected active power and reactive power, line active power and reactive power, and respectively performing 1) operation on each measurement set to generate four sets containing K trees, calculating average height of each set of trees, further calculating abnormal scores of sample information in the sets, and indicating that the sample information is abnormal if the abnormal scores exceed a set threshold;
wherein the closer the anomaly score of the sample is to 1, the higher the likelihood that it is anomaly information; if the anomaly score is less than 0.5, it can be basically determined as normal data.
Specific: for each measurement set g, the average height of m samples thereinScore of abnormalityThe calculation is as follows:
(2)
(3)
wherein,，is Euler constant;
h (g) is the average of the path lengths of each sample point in the measurement set g from the root node to the node where it is located, with respect to the number of samples m;
e (h (g)) is the average of the path lengths in K trees calculated for the measurement set g with respect to the number of trees, i.e. the average of h (g) for T trees.
S2: obtaining a power predicted value at an abnormal point by adopting historical node power measurement information and a pre-trained data driving prediction model based on an LSTM neural network;
specifically, historical node power measurement information and historical node state information are obtained, wherein the historical node state information is used for subsequent state transfer functionsThe historical information is input into a data-driven prediction model based on an LSTM neural network for training, and single-variable single-step prediction is carried out to obtain the state variable prediction quantity at the current t moment. The node state information comprises node voltage amplitude and phase angle, and is specifically expressed as:
and there is->(4)
Wherein,is a node state information set;
n, T the number of nodes and the measurement period;
and->The voltage amplitude and phase angle of the node i at the time t are respectively shown.
In the aspect of data driving prediction, an LSTM neural network is built, the LSTM (Long Short Term Memory, LSTM, long-term memory) neural network is trained through historical power data of each node, missing and abnormal data are predicted, and a predicted value of load power at an abnormal point is obtained, so that model prediction is realized, namely, the neural network is trained based on the historical power data by building a long-term neural network model, and prediction correction of abnormal injection power of the node in a state variable of a power system is realized. Specific:
inputting the first historical node power measurement information set into a data driving prediction model based on an LSTM neural network for pre-training, and inputting the second historical node power measurement information set into the pre-trained data driving prediction model based on the LSTM neural network to obtain a power prediction value at an abnormal point;
in specific implementation, S1 may select to use 2010 historical data, S2 uses 2010 historical data as the first set of historical node power measurement information for model pre-training, uses 2011 historical data for prediction, and S3 uses 2010 historical data.
Z history measurement data in equations (5) - (10) represent input data, and typically o of the last module represents the output anomaly information prediction value, i.e., the power prediction value at the anomaly, and equations (5) - (10) may be applied to 、、And->The four values, i.e. the quantity measurement at the current instant t +.>And (5) predicting.
When the data driving prediction model based on the LSTM neural network is pre-trained, the following formula is followed by the information transmission in the neurons of the LSTM neural network:
(5)
(6)
(7)
(8)
(9)
(10)
wherein,representing a forget gate, an input gate and an output gate respectively;
、、the weights of the forget gate, the input gate and the output gate are respectively represented;
、、、gate control unit bias values respectively representing states of a forgetting gate, an input gate, an output gate and a neural unit;
an input vector representing an LSTM cell;
、the states of neurons at the previous moment and the current moment are represented respectively;
representing the state of the input node at the current time t;
、the hidden layer state variables at the previous moment and the current moment are represented respectively;
and Tanh represent Sigmoid and Tanh functions, respectively.
S3: carrying out real-time classified prediction on the load based on a model driving prediction model to obtain a power prediction value at an abnormal point;
in the aspect of model driving prediction, real-time classified prediction of the load is carried out by combining a load model coefficient curve after user classification, and the active power and the reactive power of the total load at the abnormal point are obtained.
The model driving prediction model has the relations of formulas (11) - (12) for the injection active power of the node i where the abnormal information at the moment t is located.
And step S3, constructing a real-time classification model prediction model for loading in combination with the user classification curve. The process can be expressed as follows:
calculating a downstream abnormal information node at t moment according to load model coefficients (Load Model Factor, LMF) by telemetering power flow data of an upstream substation node rThe power demand belonging to the class p load is calculated as follows:
(11)
wherein,the total node number served by the node r of the upstream substation of the node i where the abnormal information at the moment t is located;
active power of a load with the type p of the node i where the abnormal information is located at the moment t is shown; the time t herein refers to any time, and the physical expression reacts in a constant relationship regardless of the specific time t.
The telemetering power of a node r of the transformer substation upstream of the node where the abnormal information is located at the moment t is shown;
a load model coefficient representing a load of type p at time t;
representing the mean value of the coefficient of the load model, which mean value represents the daily variation of the typical load, the value of the specific mean value at time t being obtained by looking up an improved daily load variation curve of different types of load at time t, which improved daily load curve is derived from statistical analysis based on historical data, typically as a function of season, temperature, days of use in the week (this function is an improved daily load curve), i.e. the curve is related to season, temperature, days of use in the week, how this is related in particular and not Described by a specific function, only the macro-level prediction correlations. In actual calculation, no pair is neededCalculating, directly obtaining by referring to improved daily load change curves of different types of loads at t time. Because it is an improved curve, the abscissa of the correspondent curve is time t, and the ordinate is E [ LMF ]]。
Wherein the load model coefficients can be derived from a profile of the daily load power variation of a certain type of load, which is typically derived by looking at literature or statistically relevant historical data.
Representing the consumption requirement of the load with the type p at the node i, wherein the consumption requirement is the number of days of the electricity/charging cycle per month;
the total predicted power at the node i where the abnormality information is located is the sum of all types of load power, and is expressed as follows:
(12)
wherein,injecting active power into a node i where abnormal information at the moment t is located;
c represents the number of load types of the node i where the anomaly information is located.
Inputting the related load information of the node where the abnormal information is located into the model-driven prediction model to realize real-time classified prediction of the load, thereby obtaining an abnormal information predicted value；
Based on the power factor of the corresponding load type obtained by experience or history statistics, calculating abnormal information The calculation formula is as follows:
wherein,injecting reactive power for the i node at the t moment;
active power of a load with the type p of the node i where the abnormal information is located at the moment t is shown;
c represents the number of load types of the node i where the abnormal information is located;
the load type at the node i ispIs a power factor of (c).
Finally, obtaining abnormal branch power information through load flow calculationThe specific formula is:
in the formula, 1jRepresenting imaginary number and having，，，；
And->The active power and the reactive power are respectively injected into the node i at the moment t;
and->Lines are respectively->Is a conductivity and susceptance of (a);
and->The voltage amplitude and the phase angle of the node i at the moment t are respectively;
and->Respectively represent t time line->Active power and reactive power of (a).
S4: learning the predicted values obtained by the S2 and the S3 by combining the historical node power measurement information to obtain a final power predicted result at the abnormal point;
constructing an integrated learning model, which comprises a first-layer basic learner and a second-layer element learner;
the integrated learning model adopts the data and the historical node power measurement information obtained in the step S2 and the step S3 to train the base learner of the first layer, takes the output of the base learner as the input of the element learner to train, and finally predicts the final result of the output of the element learner, and the method is specific:
Respectively adopting the power predicted values at the abnormal points obtained by the S2 and the S3 to replace the abnormal information in the S1 real-time node power measurement information, combining the historical node power measurement information (the second historical node power measurement information set) obtained in the S2 to form two groups of training sets, namely, replacing the abnormal information in the S1 real-time node power measurement information by the power predicted values at the abnormal points obtained by the S2, and combining the historical node power measurement information obtained in the S2 to form a group of training sets; s3, replacing the abnormal information in the S1 real-time node power measurement information by the power predicted value at the abnormal point, forming another large group of training sets by combining the historical node power measurement information obtained in the S2, and then carrying out the following prediction operation on each large group of training sets:
to avoid overfitting, the model is typically trained in a cross-validation manner at the base learner stage, i.e., the training set is divided into several subgroups for training the multiple base learners of the first layer. Training corresponding base learners respectively for each group of data, and outputting respective prediction results by each base learner; in the second layer element learner, the output of all the first layer element learners is used as the input of the element learner to train, and the output of the element learner is used as the final prediction result, namely the overall prediction result based on the model and the data model. The training set is divided into a plurality of groups for training a plurality of basic learners, wherein the number of the groups is the same as that of the basic learners, and the training set is randomly divided into 5 non-overlapping groups with basically equal size by a common 5-fold cross verification method; training the corresponding basic learners by adopting each group of data, and outputting respective prediction results by each basic learner; the outputs of all the base learners are trained as inputs to the meta-learner, the outputs of the meta-learner being the power predictions at the outliers. The multiple predicted results obtained by the two training sets are input into a meta learner of a second layer to obtain one predicted result, namely the final predicted result.
S5: and taking the final power prediction result as the pseudo-measurement of the current moment, replacing the abnormal information of the current moment, and carrying out state estimation by adopting all node power measurement information of the current moment after the abnormal information is replaced, so as to realize dynamic state estimation based on model data dual-drive.
And taking all measured information at the current moment after the abnormal information is replaced as system input to perform state estimation, and adopting a particle filtering method in the process to be more fit with the actual distribution network noise so as to realize more accurate state estimation of the power system.
Further preferably, all node power measurement information at the current moment after the abnormal information is replaced is adopted for solving the problem that the measurement data of the actual power distribution network is interfered by non-Gaussian noiseThe model for performing the state estimation is:
(13)
(14)
wherein,the state estimate at time t, which is herein indicated at any time, is a unified, constant physical constraint, which can be understood in a narrow sense herein as the predicted time.
A state transfer function representing a time t, and the process of the state transfer can be expressed as: inputting historical state data at the previous t-1 moment into LSTM for training, and performing single-step single-variable prediction to obtain the state variable prediction quantity at the current t moment;
，Respectively representing process noise and measurement noise, wherein the process noise can be Cauchy distributed noise, and the measurement noise can be Laplacian noise in the embodiment; process noise refers to noise of the design system itself;
a nonlinear mapping function of the state quantity is measured for the quantity.
Nonlinear equation from measurement to state quantity using particle filteringThe specific procedure is represented as follows:
for a pair ofAs an important issue for state estimation, the modeling of (c) is here calculated in a particle filtering manner. The main idea is that the probability density function of the system random variable is simulated by discrete random sampling points, and the result is averaged to finally obtain the minimum variance estimation of the state. Firstly, it should be clear that for formula (14), the known amount is +.>And the state quantity is needed to be solved>。
Second, it should be noted thatIs->Described as a mapping relationshipIs difficult to describe in mathematical language, and can be understood as a process, in the present invention +.>It can be understood that a black box process (the real state quantity data at the time t is input, and the output is obtained by the mapping and the noise is superimposed>) However, the inverse transformation of the process, i.e., knowing the amount of superimposed noise, performed by the present invention is actually how to determine the actual state quantity, and the specific process is the following processes (1) - (4).
Meanwhile, for the particle filtering method, the result of the state estimation can be usedThe representation is the requirement。
The non-linear mapping logic from measurement to state quantity is:
the state posterior probability is adopted to represent a state estimation result, and the posterior probability is obtained through discrete sample points obtained through random sampling, so that the state estimation result represented by a particle weight polynomial is obtained; solving the particle weight, removing the items with the particle weight smaller than the set value, and summing the rest items to obtain a final state estimation result. The specific process is (1) - (4):
(1) The state estimation result of the system can be represented by the state posterior probability of the system, namely, a set of weighted samples is approximately represented as follows:
(15)
wherein,w is W samples sampled in posterior probability,>status information at 0:T +.>Time measurement information 1:T->Refers to the importance probability density function, +.>Refers to the w-th sample of the posterior probability,is->The process is that the state information at the moment 0:T is input into LSTM to obtain the state output at the current moment;
(2) In practical application, it is difficult to directly follow the posterior probabilitySamples are extracted from the distribution, for this purpose Bayesian importance sampling is used, i.e. a simple, easy-to-sample proposed distribution is introduced +. >I.e. random sampling to obtain discrete sample points to approximately obtain a posterior probability, then the function +.>Desired rewriteability of (1) as
(16)
Wherein,representing the weight of each particle, which is calculatedThe solution process is as follows (3).
(3) Solving for the weight of the particle in equation (16):
(3.1) first, will \The significance probability density function of (c) is decomposed in the following form.
(17)
(3.2) the recursive form of the posterior probability density function can be expressed as follows:
(18)/>
it can be seen that there is a timing relationship between the final right side equation and the left side;
(3.3) substituting the formula (17) and the formula (18) into the formula of the particle weight in the formula (2), the recursive form of the particle weight can be expressed as:
(19)
(4) Copying the particles with large weight by a polynomial resampling method, deleting the particles with small weight, and finally obtainingI.e. minimum variance estimation of system state +.>. Specific: according to formula (19)Calculating to obtain +.>The +.>Items smaller than the set valueRemoving the remaining items and summing them to obtain +.>，Namely +.>That is, the state estimation obtained in S5 is also the state estimation value obtained in the present invention.
Application examples:
fig. 2 is a standard IEEE33 node power distribution network system, with a reference voltage of 12.66KV, and a state estimation of the power distribution network system is performed, which includes the steps of:
1) Load data in 2010 is selected as a measurement training set, lappas noise is superimposed on all measurement data, the parameters of the noise are 0 as the mean value, and the variance matrix isAbnormal data at the current time is detected according to formulas (3) - (4).
2) According to the internal relation of LSTM, referring to formulas (5) - (10), load data in 2010 is used as a training set to be input into LSTM for training, and measurement in 2011 is used as a test set to be input into LSTM for prediction, so that power prediction at an abnormal point under data driving is calculated;
generating a daily load curve according to 2010 historical data to obtain a load model coefficientSubstituting the average daily curve into formulas (11) - (12), and calculating power prediction at an abnormal point under the driving of a model;
and combining the data and the prediction result of the model, obtaining the final predicted power at the abnormal point through ensemble learning, and replacing the abnormal data by using the final predicted power data as pseudo-quantity measurement.
3) Performing state estimation on the power distribution network, referring to formulas (15) - (19) in a specific process, and outputting a final result of the state estimation。
In the example, the root mean square Error (Root Mean Squared Error, RMSE) and the Absolute Error (AE) are used to evaluate the result of the state Error.
Fig. 3 shows the filtering effect of the node 7 and the node 23 under the laplace noise, and it can be seen that different noises have little influence on the final state estimation by the method of the present invention, and the calculation result is almost the same as the actual result. Meanwhile, the estimated performance analysis under the influence of noise is shown in Table 1, in which it can be seen that the AE value of the node is locatedMagnitude of RMSE is also +.>The order of magnitude, the accuracy of the state estimation is very high, can satisfy the requirement of actual system.
TABLE 1 estimation results of the proposed method under the influence of Laplace noise
In order to further illustrate the accuracy of the method provided by the invention when abnormal data are contained, the following situations are set as comparison: assuming that the measurement information of the nodes 7, 20,23 is randomly missing, the error from sample sequence 9 to sample sequence 23 is 20%. And comparing the proposed method with UKF (Unscented Kalman Filter) algorithm, the result is shown in fig. 4.
As can be seen in fig. 4, the abnormal adjustment has an effect on AE of each algorithm, the AE fluctuation of the UKF is obvious and the fluctuation amplitude is far greater than the proposed state estimation, and the fluctuation of the proposed algorithm is also very small, and the accuracy and the robustness are far better than those of the UKF algorithm.
The embodiment 2 of the invention provides a dynamic state estimation system based on model data double driving, which comprises:
The information acquisition and anomaly identification module is used for acquiring real-time node power measurement information of the power system, carrying out anomaly score calculation on the real-time node power measurement information by combining the historical node power measurement information, and identifying the current moment anomaly information, wherein the corresponding node is an anomaly point;
the data driving prediction module is used for obtaining a power prediction value at an abnormal point by adopting historical node power measurement information and a pre-trained data driving prediction model based on the LSTM neural network;
the model driving prediction module is used for carrying out real-time classified prediction on the load based on the model driving prediction model to obtain a power prediction value at an abnormal point;
the integrated learning module is used for learning the predicted value obtained by the data driving prediction module and the model driving prediction module by combining the historical node power measurement information to obtain a final power predicted result at an abnormal point;
and the state estimation module is used for taking the final power prediction result as the pseudo-quantity measurement of the current moment, replacing the abnormal information of the current moment, and carrying out state estimation by adopting all node power measurement information of the current moment after the abnormal information is replaced, so as to realize the dynamic state estimation based on the model data double driving.
A terminal comprising a processor and a storage medium; the storage medium is used for storing instructions;
the processor is operative to perform the steps of the method of embodiment 1 in accordance with the instructions.
A computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of the method described in embodiment 1.
The invention has the beneficial effects that compared with the prior art:
1) According to the method, the problem that the distribution network measurement data is easy to be abnormal is considered, the abnormal score calculation is carried out on the real-time node power measurement information by combining the historical node power measurement information so as to detect the missing data and the error data, and the reliability of the input data is improved;
2) According to the invention, the data driving prediction based on the LSTM neural network, the model driving prediction and the integrated learning are combined to realize the model data mixed prediction, the node power is predicted from two parts of data trend and actual physical meaning, and finally, the integrated learning obtains better pseudo-quantity measurement, so that the influence of bad data on a state estimation result is reduced, and the accuracy of the state estimation measurement is improved;
3) According to the invention, the condition that the traditional Kalman filtering method is limited by the noise model, the measurement data of the power distribution network under the non-Gaussian noise interference is difficult to process is considered, the particle filtering is introduced to process the noise, and the condition estimation is carried out based on the particle filtering, so that the noise distribution is not limited by the model, and the actual power system condition can be reflected more.
The present disclosure may be a system, method, and/or computer program product. The computer program product may include a computer readable storage medium having computer readable program instructions embodied thereon for causing a processor to implement aspects of the present disclosure.
The computer readable storage medium may be a tangible device that can hold and store instructions for use by an instruction execution device. The computer readable storage medium may be, for example, but not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium would include the following: portable computer disks, hard disks, random Access Memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static Random Access Memory (SRAM), portable compact disk read-only memory (CD-ROM), digital Versatile Disks (DVD), memory sticks, floppy disks, mechanical coding devices, punch cards or in-groove structures such as punch cards or grooves having instructions stored thereon, and any suitable combination of the foregoing. Computer-readable storage media, as used herein, are not to be construed as transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., optical pulses through fiber optic cables), or electrical signals transmitted through wires.
The computer readable program instructions described herein may be downloaded from a computer readable storage medium to a respective computing/processing device or to an external computer or external storage device over a network, such as the internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmissions, wireless transmissions, routers, firewalls, switches, gateway computers and/or edge servers. The network interface card or network interface in each computing/processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium in the respective computing/processing device.
Computer program instructions for performing the operations of the present disclosure can be assembly instructions, instruction Set Architecture (ISA) instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, c++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer readable program instructions may be executed entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computer (for example, through the Internet using an Internet service provider). In some embodiments, aspects of the present disclosure are implemented by personalizing electronic circuitry, such as programmable logic circuitry, field Programmable Gate Arrays (FPGAs), or Programmable Logic Arrays (PLAs), with state information of computer readable program instructions, which can execute the computer readable program instructions.
Finally, 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 above embodiments, it should be understood by those skilled in the art that: modifications and equivalents may be made to the specific embodiments of the invention without departing from the spirit and scope of the invention, which is intended to be covered by the claims.
Claims (13)
1. A dynamic state estimation method based on model data double driving is characterized in that:
the method comprises the following steps:
s1: acquiring real-time node power measurement information of the power system, carrying out anomaly score calculation on the real-time node power measurement information by combining the historical node power measurement information, and identifying the current moment anomaly information, wherein the corresponding node is an anomaly point;
s2: obtaining a power predicted value at an abnormal point by adopting historical node power measurement information and a pre-trained data driving prediction model based on an LSTM neural network;
s3: carrying out real-time classified prediction on the load based on a model driving prediction model to obtain a power prediction value at an abnormal point;
s3, a model driving prediction model is adopted, and an injection active power predicted value at an abnormal point is obtained according to the load type of the node where the abnormal information is located;
The injected active power predictions at the outliers are:
(12)
(11)
wherein,injecting active power into a node i where abnormal information at the moment t is located;
c represents the number of load types of the node i where the abnormal information is located;
the total node number served by the node r of the upstream substation of the node i where the abnormal information at the moment t is located;
active power of a load with the type p of the node i where the abnormal information is located at the moment t is shown;
the telemetering power of a node r of the transformer substation upstream of the node where the abnormal information is located at the moment t is shown;
a load model coefficient representing a load of type p at time t;
representing the average value of the load model coefficients;
representing the consumption demand of a load of type p at node i;
s4: learning the predicted values obtained by the S2 and the S3 by combining the historical node power measurement information to obtain a final power predicted result at the abnormal point;
s5: and taking the final power prediction result as the pseudo-measurement of the current moment, replacing the abnormal information of the current moment, and carrying out state estimation by adopting all node power measurement information of the current moment after the abnormal information is replaced, so as to realize dynamic state estimation based on model data dual-drive.
2. The method for dynamic state estimation based on model data dual driving according to claim 1, wherein:
In S1, the node power measurement information includes node injection power and line power, where the node injection power includes injection active power and reactive power of the node, and the line power includes active power and reactive power of the line.
3. The method for dynamic state estimation based on model data dual driving according to claim 1, wherein:
in S1, the process of identifying the anomaly information includes:
1) Dividing the power measurement information of the historical node into four measurement sets of active power and reactive power which are injected, active power and reactive power of a line, randomly selecting M information samples from each set, taking the M information samples as root nodes of a tree, randomly generating cutting points, dividing the samples into two parts by taking the information size of the cutting points as a standard, placing the two parts on two sides of the cutting points, and repeatedly cutting in the sub-nodes until one information data is finally left or the sub-nodes reach a limited height; generating cutting points and repeating cutting operation circularly and randomly to generate K trees;
2) And calculating the average height of K trees generated by each set, further calculating the abnormal score of the sample information in the set, and indicating that the sample information is abnormal if the abnormal score exceeds a set threshold.
4. A method for dynamic state estimation based on model data dual driving according to claim 3, wherein:
for measurement set g, its sample information anomaly scoreThe method comprises the following steps:
(3)
(2)
wherein h (g) is the average of the path length of each sample point in the tree corresponding to the measurement set g from the root node to the node where the sample point is located with respect to the sample number m;
e (h (g)) is the average value of h (g) of K trees;
average height for m samples;
for intermediate calculation parameters +.>Is Euler constant.
5. The method for dynamic state estimation based on model data dual driving according to claim 1, wherein:
in S2, when the data-driven prediction model based on the LSTM neural network is pre-trained, the following formula is followed by the intra-neuron information transfer of the LSTM neural network:
(5)
(6)
(7)
(8)
(9)
(10)
wherein,representing a forget gate, an input gate and an output gate respectively;
、、the weights of the forget gate, the input gate and the output gate are respectively represented;
、、、respectively representing forgetful doorThe gate control unit bias of the input gate, the output gate and the neural unit state;
an input vector representing an LSTM cell;
、the states of neurons at the previous moment and the current moment are represented respectively;
representing the state of the input node at the current time t;
、The hidden layer state variables at the previous moment and the current moment are represented respectively;
and Tanh represent Sigmoid and Tanh functions, respectively.
6. The method for dynamic state estimation based on model data dual driving according to claim 1, wherein:
and S3, calculating a predicted value of the reactive power injected into the node where the abnormal information is located on the basis of the power factor, and calculating the predicted value of the branch power information where the abnormal point is located through the power flow.
7. The method for dynamic state estimation based on model data dual driving of claim 6, wherein:
the calculation formula of the predicted value of the injected reactive power of the node where the abnormal information is located is as follows:
wherein,injecting reactive power into a node i where abnormal information is located at the moment t;
active power of a load with the type p of the node i where the abnormal information is located at the moment t is shown;
c represents the number of load types of the node i where the abnormal information is located;
the load type at the node i ispIs a power factor of (c).
8. The method for dynamic state estimation based on model data dual driving according to claim 1, wherein:
s4, the concrete process is as follows:
the power predicted values at the abnormal points obtained by the S2 and the S3 are adopted to replace the abnormal information in the S1 real-time node power measurement information, two large groups of training sets are formed by combining the historical node power measurement information in the S2, and the following prediction operation is carried out on each large group of training sets: dividing the training set into a plurality of subgroups for training a plurality of base learners, wherein the number of subgroups is the same as the number of base learners; training the corresponding basic learners by adopting each group of data, and outputting respective prediction results by each basic learner;
The outputs of all the base learners are trained as inputs to the meta-learner, the outputs of the meta-learner being the power predictions at the outliers.
9. The method for dynamic state estimation based on model data dual driving according to claim 1, wherein:
s5, adopting all node power measurement information at the current moment after the abnormal information is replacedThe model for performing the state estimation is:
(13)
(14)
wherein,a state estimation value representing the time t;
a state transfer function representing time t;
，respectively representing process noise and measurement noise;
a nonlinear mapping function for measuring a state quantity based on the quantity of the particle filter.
10. The method for dynamic state estimation based on model data dual driving according to claim 9, wherein:
the non-linear mapping logic from measurement to state quantity is:
the state posterior probability is adopted to represent a state estimation result, and the posterior probability is obtained through discrete sample points obtained through random sampling, so that the state estimation result represented by a particle weight polynomial is obtained; solving the particle weight, removing the items with the particle weight smaller than the set value, and summing the rest items to obtain a final state estimation result.
11. A model data dual-drive based dynamic state estimation system for implementing the method of any one of claims 1-10, characterized by: the dynamic state estimation system includes:
the information acquisition and anomaly identification module is used for acquiring real-time node power measurement information of the power system, carrying out anomaly score calculation on the real-time node power measurement information by combining the historical node power measurement information, and identifying the current moment anomaly information, wherein the corresponding node is an anomaly point;
the data driving prediction module is used for obtaining a power prediction value at an abnormal point by adopting historical node power measurement information and a pre-trained data driving prediction model based on the LSTM neural network;
the model driving prediction module is used for carrying out real-time classified prediction on the load based on the model driving prediction model to obtain a power prediction value at an abnormal point;
the integrated learning module is used for learning the predicted value obtained by the data driving prediction module and the model driving prediction module by combining the historical node power measurement information to obtain a final power predicted result at an abnormal point;
and the state estimation module is used for taking the final power prediction result as the pseudo-quantity measurement of the current moment, replacing the abnormal information of the current moment, and carrying out state estimation by adopting all node power measurement information of the current moment after the abnormal information is replaced, so as to realize the dynamic state estimation based on the model data double driving.
12. A terminal comprising a processor and a storage medium; the method is characterized in that:
the storage medium is used for storing instructions;
the processor being operative according to the instructions to perform the steps of the method according to any one of claims 1-10.
13. Computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of the method according to any of claims 1-10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310999963.6A CN116742623B (en) | 2023-08-10 | 2023-08-10 | Dynamic state estimation method and system based on model data dual-drive |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310999963.6A CN116742623B (en) | 2023-08-10 | 2023-08-10 | Dynamic state estimation method and system based on model data dual-drive |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116742623A CN116742623A (en) | 2023-09-12 |
| CN116742623B true CN116742623B (en) | 2023-11-07 |
Family
ID=87917181
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310999963.6A Active CN116742623B (en) | 2023-08-10 | 2023-08-10 | Dynamic state estimation method and system based on model data dual-drive |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116742623B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112884237A (en) * | 2021-03-11 | 2021-06-01 | 山东科技大学 | Power distribution network prediction auxiliary state estimation method and system |
| CN115063058A (en) * | 2022-08-19 | 2022-09-16 | 东方电子股份有限公司 | Comprehensive energy situation perception system based on model driving and data driving |
-
2023
- 2023-08-10 CN CN202310999963.6A patent/CN116742623B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112884237A (en) * | 2021-03-11 | 2021-06-01 | 山东科技大学 | Power distribution network prediction auxiliary state estimation method and system |
| CN115063058A (en) * | 2022-08-19 | 2022-09-16 | 东方电子股份有限公司 | Comprehensive energy situation perception system based on model driving and data driving |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116742623A (en) | 2023-09-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Rafiq et al. | Regularized LSTM based deep learning model: first step towards real-time non-intrusive load monitoring | |
| Sangiorgio et al. | Robustness of LSTM neural networks for multi-step forecasting of chaotic time series | |
| Li et al. | Wind power forecasting using neural network ensembles with feature selection | |
| Le Son et al. | Remaining useful lifetime estimation and noisy gamma deterioration process | |
| Caceres et al. | A probabilistic Bayesian recurrent neural network for remaining useful life prognostics considering epistemic and aleatory uncertainties | |
| Li et al. | Missing traffic data: comparison of imputation methods | |
| Du et al. | Probabilistic time series forecasting with deep non‐linear state space models | |
| Wanqing et al. | Multifractional Brownian Motion and Quantum‐Behaved Partial Swarm Optimization for Bearing Degradation Forecasting | |
| Yang et al. | Robust global identification of linear parameter varying systems with generalised expectation–maximisation algorithm | |
| Rasul et al. | Lag-llama: Towards foundation models for probabilistic time series forecasting | |
| Tobar et al. | Unsupervised state-space modeling using reproducing kernels | |
| Lee et al. | Detecting anomalies in meteorological data using support vector regression | |
| Xing et al. | Comparative study of reformed neural network based short‐term wind power forecasting models | |
| CN116796639A (en) | Short-term power load prediction method, device and equipment | |
| Xiaohui et al. | A Novel Step‐by‐Step High‐Order Extended Kalman Filter Design for a Class of Complex Systems with Multiple Basic Multipliers | |
| Biggio et al. | Ageing-aware battery discharge prediction with deep learning | |
| Li et al. | Kullback–Leibler divergence for interacting multiple model estimation with random matrices | |
| CN110490378A (en) | The calculation method of Power Network Status Estimation precision based on cloud SCADA big data | |
| CN116742623B (en) | Dynamic state estimation method and system based on model data dual-drive | |
| Xu et al. | Deep‐learning‐based scenario generation strategy considering correlation between multiple wind farms | |
| Domański et al. | Outliers in Control Engineering: Fractional Calculus Perspective | |
| Fuller et al. | Probabilistic inversion of expert assessments to inform projections about Antarctic ice sheet responses | |
| Ma et al. | Parameter estimation of markov‐switching hammerstein systems using the variational bayesian approach | |
| Tran et al. | Effects of Data Standardization on Hyperparameter Optimization with the Grid Search Algorithm Based on Deep Learning: A Case Study of Electric Load Forecasting | |
| Liu et al. | Short-term power load forecasting during the COVID-19 pandemic using XGBoost and Copula theory |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |