CN116011657B - Optimization method, device and system for power distribution network load prediction model based on miniature PMU - Google Patents
Optimization method, device and system for power distribution network load prediction model based on miniature PMU Download PDFInfo
- Publication number
- CN116011657B CN116011657B CN202310043210.8A CN202310043210A CN116011657B CN 116011657 B CN116011657 B CN 116011657B CN 202310043210 A CN202310043210 A CN 202310043210A CN 116011657 B CN116011657 B CN 116011657B
- Authority
- CN
- China
- Prior art keywords
- load prediction
- task
- model
- data
- training
- 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
- 238000009826 distribution Methods 0.000 title claims abstract description 45
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000005457 optimization Methods 0.000 title claims abstract description 12
- 238000012549 training Methods 0.000 claims abstract description 65
- 238000012545 processing Methods 0.000 claims abstract description 11
- 238000012360 testing method Methods 0.000 claims description 23
- 238000004364 calculation method Methods 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 6
- 238000005259 measurement Methods 0.000 description 13
- 241000728173 Sarima Species 0.000 description 7
- 210000004027 cell Anatomy 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 7
- 230000001360 synchronised effect Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 238000007637 random forest analysis Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 101100004280 Caenorhabditis elegans best-2 gene Proteins 0.000 description 1
- 244000068645 Carya illinoensis Species 0.000 description 1
- 235000009025 Carya illinoensis Nutrition 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000004146 energy storage Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000036316 preload Effects 0.000 description 1
- 238000011946 reduction process Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 230000001932 seasonal effect Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
- Y02E40/70—Smart grids as climate change mitigation technology in the energy generation sector
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the technical field of situation awareness, in particular to a power distribution network load prediction model optimization method, device and system based on a miniature PMU. The method comprises the steps of obtaining task data of a sample load prediction task and a feature set F of the sample load prediction task; training a plurality of load prediction models according to task data of sample load prediction tasks, and acquiring a corresponding optimal load prediction model for solving each sample load prediction task based on root mean square error to form an optimal load prediction model set phi; the feature set F and the optimal load prediction model set phi are combined to form metadata<F,Φ>The method comprises the steps of carrying out a first treatment on the surface of the Utilizing metadata<F,Φ>Training the plurality of meta learners respectively to obtain a plurality of trained meta learners; processing the load prediction task feature set by using a plurality of trained meta learners to obtain a plurality of model recommendation result data
Recommending multiple models to result data
Obtaining single model recommendation result data through a voter
Description
Technical Field
The invention relates to the technical field of situation awareness, in particular to a power distribution network load prediction model optimization method, device and system based on a miniature PMU.
Background
The power system load prediction is an important component of situation awareness and is also one of basic support technologies for system scheduling operation. The synchronous phasor measurement unit (Phasor Measurement Unit, PMU) is a phasor measurement unit formed by taking GPS global positioning system second pulse as a synchronous clock, can endow global measurement data in a system with a uniform time scale, can directly measure phasor information in a power distribution network, ensures the synchronism and accuracy of the measurement data, and can be used in the fields of dynamic monitoring, system protection, system analysis, prediction and the like of a power system. In recent years, as a large number of PMUs are connected into a power system, the operation and control of the power system are greatly influenced, how to utilize the miniature PMUs in a good power grid, and research on the key support technology of the next-generation distribution network automation system based on the miniature PMU technology adapting to the distribution network requirements has become a major scientific proposition in the fields of energy and power systems.
With the large-scale access of the distributed power supply, the current operation mode of the power distribution network is more flexible. For example, a building equipped with photovoltaic and energy storage systems may be a sales producer to achieve self-sufficiency of electrical energy; part of the source-containing network in the system can be independently operated as a micro-grid off-grid; the network topology of the system may implement dynamic reconfiguration, etc., as desired. To meet these diverse operating requirements, corresponding load prediction techniques are required to provide support thereto. Fig. 1 depicts the different load prediction tasks in a power distribution network from five dimensions of data resolution, load level, impact factors, prediction window, historical data length. For example, where red pentagons are combined together, a single user's daily preload prediction task is defined with a data resolution of 1 hour, a usable history data length of 1 year, and modeling can be performed using temperature as an influencing element.
As can be seen from fig. 1, these heterogeneous predictive tasks differ significantly in terms of data characteristics, predictive requirements, etc.
In 1995, d.h.wolpert et al proposed no free lunch theorem (No Free Lunch Theorem): any predictive function, if performing well on some training samples, must perform poorly on other training samples; if there is no assumption about the a priori distribution of the data in the feature space, it performs as well as it does not. According to "No Free Lunch Theory", these heterogeneous predictive tasks are difficult to solve well with the same predictive model.
The Long short-term memory (LSTM) is a special RNN, and the RNN cells in the hidden layer are replaced by LSTM cells, so that the Long-term memory is realized. Through continuous evolution, the LSTM cell structure which is most widely used at present is shown in fig. 2, z is an input module, and the forward calculation method can be expressed as
i t ＝σ(W xi x t +W hi h t-1 +W ci c t-1 +b i ) (1)
f t ＝σ(W xf x t +W hf h t-1 +W cf c t-1 +b f ) (2)
c t ＝f t c t-1 +i t tanh(W xc x t +W hc h t-1 +b c ) (3)
o t ＝σ(W xo x t +W ho h t-1 +W co c t +b o ) (4)
h t ＝o t tanh(c t ) (5)
Wherein i, f, c, o is an input gate, a forgetting gate, a cell state and an output gate respectively; w and b are respectively corresponding weight coefficient matrixes and bias items; sigma and tanh are sigmoid and hyperbolic tangent activation functions, respectively.
The LSTM model training process can be roughly divided into four steps, namely calculating the output value of the LSTM cells according to a forward calculation method from the formula (1) to the formula (5); calculating the error term of each LSTM cell reversely, wherein the error term comprises 2 reverse propagation directions of time and network level; calculating the gradient of each weight according to the corresponding error term; the weights are updated using a gradient-based optimization algorithm.
LSTM networks are well suited for classification, processing and prediction based on time series data, but are not necessarily suitable for all load prediction tasks in a power distribution network. The LSTM model is characterized in that the number of training samples required by the LSTM model is large, the calculation time is long, and the LSTM model is difficult to be suitable for small samples and prediction scenes with high real-time requirements. In addition, part of the prediction task history load sequence is smoother, and can be solved by a more concise prediction model without introducing an LSTM model. The above examples demonstrate that LSTM is not a universal prediction method capable of solving heterogeneous prediction tasks in a power distribution network. The model provided by the method is preferably constructed, the optimal prediction model can be selected according to different prediction tasks, and the adaptation problem of the model is well solved.
Disclosure of Invention
In order to solve the technical problems in the prior art, the invention provides a power distribution network load prediction model optimization method, device and system based on a miniature PMU.
In order to achieve the above object, the embodiment of the present invention provides the following technical solutions:
in a first aspect, in one embodiment provided by the present invention, a method for optimizing a load prediction model of a power distribution network based on a micro PMU is provided, the method comprising the steps of:
acquiring task data of a sample load prediction task and a feature set F of the sample load prediction task;
training a plurality of load prediction models according to the task data of the sample load prediction task, and acquiring a corresponding optimal load prediction model for solving each sample load prediction task based on root mean square error to form an optimal load prediction model set phi;
the feature set F and the optimal load prediction model set phi together form metadata < F, phi >; training the plurality of element learners respectively by utilizing the metadata < F, phi > to obtain a plurality of trained element learners;
processing the load prediction task feature set by using a plurality of trained meta learners to obtain a plurality of model recommendation result data
Recommending result data of said plurality of models +.>
Obtaining single model recommended result data by a voter>
As a further aspect of the present invention, the sample load prediction task data includes J pairs of data samples<X j ,y j >Wherein X is j Is the input data of the load prediction model, and the dimension is N j ×M j ；y j Is a true load value with dimension N j X 1, where J e [1 ], J]。
As a further aspect of the present invention, the X j Included
And->
y j Comprises->
And->
Wherein the said
By the following formulaAnd (3) calculating:
wherein the method comprises the steps of
Represents the mth attribute of the nth sample. For the j-th load prediction task, a load prediction model i is used for B The prediction result is marked as->
As a further aspect of the present invention, the feature set F has dimensions of jxd.
As a further aspect of the present invention, the load prediction model performs the following formula calculation:
wherein the method comprises the steps of
To measure the loss function of the distance between the predicted and real results, θ * Then it is the load prediction model i B Is described.
As a further aspect of the present invention, the metadata<F,Φ>Dividing into metadata training sets<F train ,Φ train >And metadata test set<F test ,Φ test >。
As a further aspect of the present invention, the training of the plurality of meta learners using the metadata < F, Φ > to obtain a trained meta learner further includes:
multiple stations in training processThe meta learner generates a plurality of corresponding training model recommendation result data
Recommending the result data of the plurality of training models
Obtaining single training model recommended result data +.>
In a second aspect, in yet another embodiment provided by the present invention, there is provided a miniature PMU-based power distribution network load prediction model optimization apparatus, the apparatus comprising: the system comprises a data acquisition module, a first training module, a second training module and an application module;
the data acquisition module is used for acquiring task data of a sample load prediction task and a feature set F of the sample load prediction task, wherein the sample load prediction task comprises J sample load prediction tasks;
the first training module is used for training a plurality of load prediction models according to the task data of the sample load prediction task, and acquiring a corresponding optimal load prediction model for solving each sample load prediction task based on root mean square error to form an optimal load prediction model set phi;
the second training module is configured to combine the feature set F and the optimal load prediction model set Φ to form metadata < F, Φ >, and train a plurality of meta learners respectively by using the metadata < F, Φ > to obtain a plurality of trained meta learners;
the application module is used for respectively processing the load prediction task feature set by utilizing a plurality of trained meta learners to obtain a plurality of model recommendation result data
Recommending result data of said plurality of models +.>
Obtaining single model recommended result data by a voter>
In a third aspect, in yet another embodiment of the present invention, a miniature PMU-based power distribution network load prediction model optimization system is provided, the system comprising: a basic learning layer, a meta learning layer and an application layer;
the basic learning layer is used for acquiring task data of a sample load prediction task, training a plurality of load prediction models according to the task data of the sample load prediction task, and acquiring a corresponding optimal load prediction model for solving each sample load prediction task based on root mean square error to form an optimal load prediction model set phi;
the meta learning layer is used for acquiring a feature set F of a sample load prediction task, and forming metadata < F, phi > together with the optimal load prediction model set phi by the feature set F; training the plurality of meta learners respectively by using the meta data < F, phi > to obtain a plurality of trained meta learners;
the application layer is used for respectively processing the load prediction task feature set by utilizing a plurality of trained meta learners to obtain a plurality of model recommendation result data
Recommending result data of said plurality of models +.>
Obtaining single model recommended result data by a voter>
The technical scheme provided by the invention has the following beneficial effects:
the invention provides the basisThe invention discloses a power distribution network load prediction model optimization method, device and system of a miniature PMU, and the method comprises the steps of obtaining task data of a sample load prediction task and a feature set F of the sample load prediction task; training a plurality of load prediction models according to the task data of the sample load prediction task, and acquiring a corresponding optimal load prediction model for solving each sample load prediction task based on root mean square error to form an optimal load prediction model set phi; the feature set F and the optimal load prediction model set phi together form metadata<F,Φ>The method comprises the steps of carrying out a first treatment on the surface of the Utilizing the metadata<F,Φ>Training the plurality of meta learners respectively to obtain a plurality of trained meta learners; processing the load prediction task feature set by using a plurality of trained meta learners to obtain a plurality of model recommendation result data
Recommending the plurality of models to result data
Obtaining single model recommended result data by a voter>
According to the invention, an element learning technology is adopted, an optimal load prediction model is recommended for each miniature synchronous phasor measurement unit (Phasor Measurement Unit, PMU) in the power distribution network, the requirement of heterogeneous load prediction tasks in the power distribution network is met, and the overall prediction accuracy is improved.
These and other aspects of the invention will be more readily apparent from the following description of the embodiments. It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention as claimed.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are necessary for the description of the embodiments or the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention and that other embodiments may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a diagram of heterogeneous load prediction tasks in a power distribution network.
FIG. 2 is a schematic diagram of the LSTM hidden layer cell structure.
Fig. 3 is a flowchart of a preferred method of a miniature PMU-based load prediction model for a power distribution network according to one embodiment of the invention.
Fig. 4 is a block diagram of a preferred apparatus for a miniature PMU-based load prediction model of a power distribution network according to an embodiment of the present invention.
Fig. 5 is a block diagram of a preferred system of a miniature PMU-based load prediction model for a power distribution network according to an embodiment of the present invention.
Fig. 6 is a voter flow of a meta-learner in a preferred system of a miniature PMU-based load prediction model for a power distribution network according to an embodiment of the present invention.
Fig. 7 is an example model selection process.
Fig. 8 is a model labeling result.
Fig. 9 is a graph of SER ratios for different prediction models.
Fig. 10 is a graph of a dimension reduction process for the feature set based on random proximity coding of T distribution.
Fig. 11 shows the test results of the meta learner at cluster 4.
Fig. 12 shows a recommendation model prediction result a.
Fig. 13 shows the recommendation model prediction result b.
In the figure: the system comprises a data acquisition module-100, a first training module-200, a second training module-300 and an application module-400.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The flow diagrams depicted in the figures are merely illustrative and not necessarily all of the elements and operations/steps are included or performed in the order described. For example, some operations/steps may be further divided, combined, or partially combined, so that the order of actual execution may be changed according to actual situations.
It is to be understood that the terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.
In particular, embodiments of the present invention are further described below with reference to the accompanying drawings.
Referring to fig. 3, fig. 3 is a flowchart of a preferred method for a miniature PMU-based power distribution network load prediction model according to an embodiment of the present invention, and as shown in fig. 3, the preferred method for a miniature PMU-based power distribution network load prediction model includes steps S10 to S40.
S10, task data of sample load prediction tasks and a feature set F of the sample load prediction tasks are obtained, wherein the sample load prediction tasks comprise J sample load prediction tasks.
In an embodiment of the present invention, the sample load prediction task data includes J pairs of data samples<X j ,y j >Wherein X is j Is the input data of the load prediction model, and the dimension is N j ×M j ；y j Is a true load value with dimension N j X 1, where J e [1 ], J]. The value range of J is a positive integer.
In an embodiment of the invention, to train the load prediction model, X is therefore j Included
And->
y j Comprises->
And->
Wherein the said
The calculation is performed by the following formula:
wherein the method comprises the steps of
Represents the mth attribute of the nth sample. For the j-th load prediction task, a load prediction model i is used for B The prediction result is marked as->
S20, training a plurality of load prediction models according to task data of the sample load prediction tasks, and acquiring a corresponding optimal load prediction model for solving each sample load prediction task based on root mean square error to form an optimal load prediction model set phi.
The number of the load prediction models is I B And each. Wherein I is B The value of (2) can be flexibly configured according to the actual demands of users. It should be noted that the selected I B The individual models should have different advantages to achieve mutual complementation so that diversified prediction tasks can be solved.
The load prediction model performs the following formula calculation:
wherein the method comprises the steps of
In order to measure the loss function of the distance between the predicted result and the real result, theta is the undetermined parameter of the predicted model, theta * Then it is the load prediction model i B Is described. When get theta through training * The prediction accuracy can then be measured by calculating the RMSE error over the test set, i.e
At all I B Among the candidate load prediction models, the model with the highest prediction accuracy for load prediction task j will be labeled Φ (j).
S30, the feature set F and the optimal load prediction model set phi together form metadata < F, phi >; and training the plurality of meta learners respectively by using the metadata < F, phi > to obtain a plurality of trained meta learners.
The dimension of the feature set F is J×D, and D is the number of features in the feature set.
For training the meta learner, the meta data<F,Φ>Dividing into metadata training sets<F train ,Φ train >And metadata test set<F test ,Φ test >。
The meta learner performs the following calculations:
wherein g w Representing a meta learner, w is a parameter to be trained of meta learning,
is a loss function of the meta learner for measuring the recommended model +.>
And the actual optimal model phi train Distance between them. The optimal parameters obtained after training w are recorded as w * . The meta learner obtains the optimal parameter w by training * Then, the precision of the test set can be tested on the test set, and the precision eta is recommended iM The calculation is performed by the following formula:
wherein K is M The task number is predicted for the load in meta-learning training.
Specifically, the element learner is essentially a multi-class machine learning model whose parameters are w (e.g., if the element learner is a neural network, then w refers to the weights of the individual neurons in the network). After training the model, w will slowly converge to the optimal value, i.e. w * 。
Since a plurality of meta learners are adopted in the invention, a plurality of model recommendation results are generated
By designing the voter according to the invention, the above recommended results can be integrated and the final recommended result +.>
The corresponding recommendation accuracy is η.
In an embodiment of the present invention, the training the plurality of meta learners using the metadata < F, Φ > to obtain a trained meta learner, further includes:
in the training process, a plurality of meta learners generate a plurality of corresponding training model recommendation result data
Recommending the result data of the plurality of training models
Obtaining single training model recommended result data +.>
The number of the meta learner is I M The value of the device can be flexibly configured according to the actual demands of users, and the minimum value is 1.
S40, respectively processing the load prediction task feature set by using a plurality of trained meta learners to obtain a plurality of model recommendation result data
Recommending result data of said plurality of models +.>
Obtaining single model recommended result data by a voter>
According to the invention, an element learning technology is adopted, an optimal load prediction model is recommended for each miniature synchronous phasor measurement unit (Phasor Measurement Unit, PMU) in the power distribution network, the requirement of heterogeneous load prediction tasks in the power distribution network is met, and the overall prediction accuracy is improved.
It should be understood that although described in a certain order, the steps are not necessarily performed sequentially in the order described. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, some steps of the present embodiment may include a plurality of steps or stages, which are not necessarily performed at the same time, but may be performed at different times, and the order of the steps or stages is not necessarily sequential, but may be performed alternately or alternately with at least a part of the steps or stages in other steps or other steps.
In one embodiment, referring to fig. 4, a preferred apparatus for a miniature PMU-based power distribution network load prediction model is also provided in an embodiment of the present invention, where the apparatus includes a data acquisition module 100, a first training module 200, a second training module 300, and an application module 400.
The data acquisition module 100 is configured to acquire task data of a sample load prediction task and a feature set F of the sample load prediction task, where the sample load prediction task includes J sample load prediction tasks.
The first training module 200 is configured to train a plurality of load prediction models according to task data of the sample load prediction task, and obtain a corresponding optimal load prediction model for solving each sample load prediction task based on root mean square error, so as to form an optimal load prediction model set Φ.
The second training module 300 is configured to combine the feature set F and the optimal load prediction model set Φ to form metadata < F, Φ >, and train the plurality of meta learners respectively by using the metadata < F, Φ > to obtain a plurality of trained meta learners.
The application module 400 is configured to process the load prediction task feature set by using a plurality of trained meta learners, respectively, to obtain a plurality of model recommendation result data
Recommending result data of said plurality of models +.>
Obtaining single model recommended result data by a voter>
According to the invention, an element learning technology is adopted, an optimal load prediction model is recommended for each miniature synchronous phasor measurement unit (Phasor Measurement Unit, PMU) in the power distribution network, the requirement of heterogeneous load prediction tasks in the power distribution network is met, and the overall prediction accuracy is improved.
In one embodiment, referring to fig. 5, a preferred system for a miniature PMU-based power distribution network load prediction model is also provided in an embodiment of the present invention, where the system includes a base learning layer, a meta-learning layer, and an application layer.
The basic learning layer is used for acquiring task data of a sample load prediction task, training a plurality of load prediction models according to the task data of the sample load prediction task, and acquiring a corresponding optimal load prediction model for solving each sample load prediction task based on root mean square error to form an optimal load prediction model set phi.
The meta learning layer is used for acquiring a feature set F of a sample load prediction task, and forming metadata < F, phi > by the feature set F and the optimal load prediction model set phi together; and training the plurality of meta learners respectively by using the metadata < F, phi > to obtain a plurality of trained meta learners.
The application layer is used for respectively processing the load prediction task feature set by utilizing a plurality of trained meta learners to obtain a plurality of model recommendation result data
Recommending result data of said plurality of models +.>
Obtaining single model recommended result data by a voter>
According to the invention, an element learning technology is adopted, an optimal load prediction model is recommended for each miniature synchronous phasor measurement unit (Phasor Measurement Unit, PMU) in the power distribution network, the requirement of heterogeneous load prediction tasks in the power distribution network is met, and the overall prediction accuracy is improved.
In an exemplary embodiment, in a specific application process of the system for optimizing a load prediction model of a power distribution network based on a micro PMU, a load prediction model, a load prediction task feature set and a load prediction task are given, and the steps are as follows:
and step S100, creating a sample load prediction task.
A large number of heterogeneous sample load prediction tasks will be created starting from five dimensions as shown in fig. 1.
And step 200, selecting four types of commonly used load prediction models as alternatives. Wherein the load prediction model comprises: autoregressive differential moving average model with seasonal features (Autoregressive Integrated Moving Average, SARIMA), LSTM model, support vector regression model (Support Vector Regression, SVR), and similarity Day model (Similar Day, SD). The four load prediction models represent four different ideas of time sequence prediction, deep learning prediction, correlation factor prediction and clustering prediction respectively. 6 different parameter structures are set for the SARIM model, and 2 different parameter structures are set for the LSTM model so that the verification element learner can select not only a large model class but also an optimal model structure. The resulting 10 candidate models are shown in table 1.
TABLE 1 alternative load prediction model
Table 1 Candidate load forecasting models
And step S300, marking a load prediction model.
In order to find the optimal predictive model phi (j) for the load predictive task j, the formulas (7) - (10) will be repeated Lj times, each time with different training and test data slicing, thus obtaining the distribution omega of the optimal load predictive model j . This process continues until a distribution Ω j Tend to stabilize. Pearson correlation coefficient P cc Can be used as termination condition, namely when omega j (L j ) And omega j (L j P between-10) cc Above 0.95, the distribution is considered to have reached a steady state. See algorithm 1 for details.
Step S400, designing a feature set containing 16 features, and quantitatively describing the load prediction task, as shown in table 2.
TABLE 2 load prediction task feature set
Table2 Feature set of forecasting tasks
Kurtosis (Kurtosis) and Skewness (Skewness) can be calculated by equations (16) and (17), where y is the historical load data sequence, N is the data sequence length, σ,
y (n) represents the nth value of the data sequence, which is the standard deviation and the mean of the historical load sequence.
Volatility (Fickleness) measures the number of times a historical sequence crosses its mean line
The correlation characteristics of the maximum autocorrelation coefficient (H-ACF) and maximum partial autocorrelation coefficient (H-PACF) load sequences are important in determining the structure of the SARIMA model. The period (Periodicity) is then directly related to the resolution of the data, e.g. for an hourly resolution payload sequence, the period is at most 24 or 168. Whereas for a sequence of one day resolution, the period is at most 30.
Step S500, a meta learner selecting and voting device.
The meta learner in the invention realizes the mapping from the task feature F to the optimal load prediction model phi (j), which is essentially a multi-classification problem. In order to improve the classification accuracy of the meta learner, the meta learner in the system comprises 4 different classifiers. The classifier comprises Random Forest (RF), K-Nearest Neighbor (KNN), naive Bayes
Bayesian, NB), and linear discriminant analysis (Linear Discrimination, LD). In order to combine the recommendation results of the classifiers and obtain a final recommendation model, the invention designs a voter based on the idea of ensemble learning, as shown in fig. 6.
The mechanism by which each meta learner implements classification is based on an internal scoring process. For example, the NB may calculate the posterior probability of each class as its score, while the RF calculates the voting results of the internally contained decision tree as its score. The meta learner iM selects as its output the model whose score is highest. A higher score means that the meta-learner has a higher confidence in its recommended results. From this property we build a score
And classification accuracy->
Mapping relation between
When a model recommendation is performed for a new load prediction task, firstly, the score of each element learner is converted into the corresponding recommendation precision according to the above formula, and a model with the highest recommendation precision is selected as a final model recommendation on the basis of the score
Results
And S600, creating a load prediction task.
Creating a large number of real, heterogeneous load prediction tasks is the basis for test element learning effectiveness. As shown in table 3, we create different load prediction tasks by combining the five dimensions shown in fig. 1.
Table 3 creates heterogeneous load prediction task Table 3Heterogeneous forecasting task creation
Resident and business load data with 15 minute and 30 minute accuracy were collected from north carolina and 1 minute accuracy from Pecan Street. The weather data for the hour accuracy was collected from the NOAA website. By combining the five dimensional features in table 3, 846 different load prediction tasks are obtained herein. These tasks have the following characteristics: 1) Mainly comprises resident and commercial loads, and industrial and agricultural loads are not considered; 2) From individual customers to the microgrid, short-term load predictions, which are most commonly used, are mainly considered to support its operation; 3) At the feeder line level, additionally considering the medium-long term load prediction task; 4) Up to 12 weather features are available; 5) 3 different historical data lengths are considered to reflect the difference in historical data availability in actual engineering applications.
And step S700, the basic learning layer acquires a model optimization result.
For the 846 load prediction tasks, an optimal prediction model of the load prediction tasks is found and marked. Fig. 7 illustrates a model selection process using a load prediction task as an example. As the number of iterations increases, the distribution of the optimal model gradually stabilizes. At 60 iterations, the correlation coefficient between it and the optimal model frequency distribution of 50 iterations reaches 0.98>0.95, so that the iteration is terminated and the model 9 with the largest frequency is selected as the optimal model for the load prediction task.
The model labeling results for all 846 load prediction tasks are shown in fig. 8. It should be noted that if the historical data of a certain load prediction task is not sufficient to train the model, the model is considered to fail and a significant prediction error is given to the model to avoid its selection.
From the results, it can be seen that model 7, i.e., LSTM (125), is the most frequently selected model. In addition, model 10, the day of similarity model, has the shortest training time, but its mean and variance of prediction errors is greater than the other models. Because the SARIMA model has high requirements on historical data, the training failure times are relatively high, and the SARIMA model is especially high-order. But when the SARIMA model can be trained, it has a better predictive performance.
To further quantify the differences between the different prediction models, a systematic error ratio (System Error Ratio, SER) is defined
Where Eselection is the RMSE of the recommended model and Ebest is the RMSE of the optimal model. The index measures the distance between the recommended model and the actual optimal model. Fig. 9 illustrates the difference in performance of different models on load prediction tasks. It can be seen that the order 2-4 models most often have similar behavior to the optimal model, i.e., SER is close to 1. However, the latter ranked model performs significantly worse than the optimal model. This indicates the necessity of developing model selection work.
And S800, evaluating the similarity of the load prediction tasks.
At the meta-learning layer, the input to the meta-learner is the feature set F of the load prediction task. We apply random proximity coding (T-distributed Stochastic Neighbor Embedding, T-SNE) based on T distribution to dimension down the feature set and draw on a two-dimensional plane as shown in fig. 10. As can be seen from the figure, the load prediction tasks are clustered into approximately 5 clusters, as shown in table 4.
TABLE 4 load prediction task cluster characterization
Table 4 Features of each LF cluster
Step S900, training and verifying results of the meta learner.
To train and test the meta learner, we split 846 load prediction tasks into training set (70%), validation set (20%), and test set (10%). Fig. 11 shows the test results of the trained meta learner on cluster 4. It can be seen that different meta-learners have each select a successful task and a failed task. By effectively integrating these, the overall model preference accuracy can be improved, as shown in table 5.
Table 5 element learner test accuracy
Table 5 Accuracy of meta learners
While recommending the optimal model, the recommendation system built herein may give the ranking of all candidate models on each load prediction task, providing a reference for the operator. As shown in table 6, it can be seen that the top-ranked 3 models all have a higher recommendation success rate and a lower SER error rate, and can be used as effective models. The recommendation system recommends a probability of 76% for an effective model.
Table 6 precision alignment of different ranking models
Table6 Accuracy comparison of LF models under different rankings
| Ordering of | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 |
| Precision of | 46％ | 17％ | 13％ | 6％ | 4％ | 3％ | 3％ | 3％ | 2％ | 3％ |
| SER | 1.14 | 1.27 | 1.34 | 1.46 | 4.18 | 2.89 | 4.48 | 3.61 | 2.61 | 3.09 |
| Number of |
0 | 0 | 2 | 10 | 10 | 12 | 12 | 17 | 14 | 11 |
Step S1000, online application testing.
After the recommendation system is trained, online application can be performed to recommend a prediction model for a new load prediction task. This section considers performing model recommendation tests on 2 new load prediction tasks, the task descriptions are shown in table 7. Task 1 is short-term load prediction at the transformer level with 30 days 15 minutes resolution historical data; mission 2 is a short term load forecast at the feeder level with historical data at a resolution of 6 months to 1 hour. According to the recommendation system, the task 1 recommendation model is SD and is a true optimal model; task 2 recommended model is SARIMA (5,1,5), true best 2 model. The prediction was performed using the recommendation model, and the results are shown in fig. 12 and 13.
Table 7 2 New test tasks
Table 7 Two testing LF tasks
It should also be understood that the term "and/or" as used in the present specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes such combinations.
It should be understood that as used herein, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context clearly supports the exception. It should also be understood that "and/or" as used herein is meant to include any and all possible combinations of one or more of the associated listed items. The foregoing embodiment of the present invention has been disclosed with reference to the number of embodiments for the purpose of description only, and does not represent the advantages or disadvantages of the embodiments.
Those of ordinary skill in the art will appreciate that: the above discussion of any embodiment is merely exemplary and is not intended to imply that the scope of the disclosure of embodiments of the invention, including the claims, is limited to such examples; combinations of features of the above embodiments or in different embodiments are also possible within the idea of an embodiment of the invention, and many other variations of the different aspects of the embodiments of the invention as described above exist, which are not provided in detail for the sake of brevity. Therefore, any omission, modification, equivalent replacement, improvement, etc. of the embodiments should be included in the protection scope of the embodiments of the present invention.
Claims (3)
1. A method for optimizing a load prediction model of a power distribution network based on a miniature PMU, the method comprising:
acquiring task data of a sample load prediction task and a feature set F of the sample load prediction task;
training a plurality of load prediction models according to the task data of the sample load prediction task, and acquiring a corresponding optimal load prediction model for solving each sample load prediction task based on root mean square error to form an optimal load prediction model set phi;
the feature set F and the optimal load prediction model set phi together form metadata < F, phi >; training the plurality of meta learners respectively by using the meta data < F, phi > to obtain a plurality of trained meta learners;
processing the load prediction task feature set by using a plurality of trained meta learners to obtain a plurality of model recommendation result data
Recommending result data of said plurality of models +.>
Obtaining single model recommended result data by a voter>
The sample load prediction task data comprises J data sample pairs<X j ,y j >Wherein X is j Is the input data of the load prediction model, and the dimension is N j ×M j ；y j Is a true load value with dimension N j X 1, where J e [1 ], J]；
The X is j Included
And->
y j Comprises->
And->
Wherein the said
By the following formulaRow calculation:
wherein the method comprises the steps of
An mth attribute representing an nth sample; for the j-th load prediction task, a load prediction model i is used for B The prediction result is marked as->
The dimension of the feature set F is J x D;
the load prediction model performs the following formula calculation:
wherein the method comprises the steps of
To measure the loss function of the distance between the predicted and real results, θ * Then it is the load prediction model i B Is defined in the set of parameters;
the metadata<F,Φ>Dividing into metadata training sets<F train ,Φ train >And metadata test set<F test ,Φ test >；
The training of the meta learner by using the meta data < F, phi > to obtain a trained meta learner further comprises:
in the training process, a plurality of meta learners generate a plurality of corresponding training model recommendation resultsData
Recommending the result data of the plurality of training models
Obtaining single training model recommended result data +.>
2. A miniature PMU-based power distribution network load prediction model optimization device employing the method of claim 1, the device comprising: the system comprises a data acquisition module, a first training module, a second training module and an application module;
the data acquisition module is used for acquiring task data of a sample load prediction task and a feature set F of the sample load prediction task, wherein the sample load prediction task comprises J sample load prediction tasks;
the first training module is used for training a plurality of load prediction models according to the task data of the sample load prediction task, and acquiring a corresponding optimal load prediction model for solving each sample load prediction task based on root mean square error to form an optimal load prediction model set phi;
the second training module is configured to combine the feature set F and the optimal load prediction model set Φ to form metadata < F, Φ >, and train a plurality of meta learners respectively by using the metadata < F, Φ > to obtain a plurality of trained meta learners;
the application module is used for respectively processing the load prediction task feature set by utilizing a plurality of trained meta learners to obtain a plurality of model recommendation result data
Recommending the plurality of modelsResult data->
Obtaining single model recommended result data by a voter>
3. A miniature PMU-based power distribution network load prediction model optimization system employing the method of claim 1, the system comprising: a basic learning layer, a meta learning layer and an application layer;
the basic learning layer is used for acquiring task data of a sample load prediction task, training a plurality of load prediction models according to the task data of the sample load prediction task, and acquiring a corresponding optimal load prediction model for solving each sample load prediction task based on root mean square error to form an optimal load prediction model set phi;
the element learning layer is used for acquiring a characteristic set F of a sample load prediction task, and forming metadata < F, phi > by the characteristic set F and the optimal load prediction model set phi together; training the plurality of element learners respectively by utilizing the metadata < F, phi > to obtain a plurality of trained element learners;
the application layer is used for respectively processing the load prediction task feature set by utilizing a plurality of trained meta learners to obtain a plurality of model recommendation result data
Recommending result data of said plurality of models +.>
Obtaining single model recommended result data by a voter>
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310043210.8A CN116011657B (en) | 2023-01-29 | 2023-01-29 | Optimization method, device and system for power distribution network load prediction model based on miniature PMU |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310043210.8A CN116011657B (en) | 2023-01-29 | 2023-01-29 | Optimization method, device and system for power distribution network load prediction model based on miniature PMU |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116011657A CN116011657A (en) | 2023-04-25 |
| CN116011657B true CN116011657B (en) | 2023-06-27 |
Family
ID=86023043
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310043210.8A Active CN116011657B (en) | 2023-01-29 | 2023-01-29 | Optimization method, device and system for power distribution network load prediction model based on miniature PMU |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116011657B (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110707763A (en) * | 2019-10-17 | 2020-01-17 | 南京理工大学 | AC/DC power distribution network load prediction method based on ensemble learning |
| CN112257868A (en) * | 2020-09-25 | 2021-01-22 | 建信金融科技有限责任公司 | Method and device for constructing and training integrated prediction model for predicting passenger flow |
| CN113159361A (en) * | 2020-12-03 | 2021-07-23 | 安徽大学 | Short-term load prediction method and system based on VDM and Stacking model fusion |
| CN113723844A (en) * | 2021-09-06 | 2021-11-30 | 东南大学 | Low-voltage transformer area theoretical line loss calculation method based on ensemble learning |
| CN114491028A (en) * | 2022-01-18 | 2022-05-13 | 四川大学 | Small sample text classification method based on regularization meta-learning |
| CN114970345A (en) * | 2022-05-25 | 2022-08-30 | 武汉大学 | Short-term load prediction model construction method, device, equipment and readable storage medium |
| CN115240871A (en) * | 2022-07-26 | 2022-10-25 | 南昌理工学院 | Epidemic disease prediction method based on deep embedded clustering element learning |
| CN115276006A (en) * | 2022-09-26 | 2022-11-01 | 江苏永鼎股份有限公司 | Load prediction method and system for power integration system |
| CN115577746A (en) * | 2022-09-15 | 2023-01-06 | 南京辰光融信技术有限公司 | Network intrusion detection method based on meta-learning |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8251704B2 (en) * | 2007-04-12 | 2012-08-28 | Microsoft Corporation | Instrumentation and schematization of learning application programs in a computerized learning environment |
| US7849043B2 (en) * | 2007-04-12 | 2010-12-07 | Microsoft Corporation | Matching educational game players in a computerized learning environment |
| US11663523B2 (en) * | 2019-09-14 | 2023-05-30 | Oracle International Corporation | Machine learning (ML) infrastructure techniques |
| US11475374B2 (en) * | 2019-09-14 | 2022-10-18 | Oracle International Corporation | Techniques for automated self-adjusting corporation-wide feature discovery and integration |
-
2023
- 2023-01-29 CN CN202310043210.8A patent/CN116011657B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110707763A (en) * | 2019-10-17 | 2020-01-17 | 南京理工大学 | AC/DC power distribution network load prediction method based on ensemble learning |
| CN112257868A (en) * | 2020-09-25 | 2021-01-22 | 建信金融科技有限责任公司 | Method and device for constructing and training integrated prediction model for predicting passenger flow |
| CN113159361A (en) * | 2020-12-03 | 2021-07-23 | 安徽大学 | Short-term load prediction method and system based on VDM and Stacking model fusion |
| CN113723844A (en) * | 2021-09-06 | 2021-11-30 | 东南大学 | Low-voltage transformer area theoretical line loss calculation method based on ensemble learning |
| CN114491028A (en) * | 2022-01-18 | 2022-05-13 | 四川大学 | Small sample text classification method based on regularization meta-learning |
| CN114970345A (en) * | 2022-05-25 | 2022-08-30 | 武汉大学 | Short-term load prediction model construction method, device, equipment and readable storage medium |
| CN115240871A (en) * | 2022-07-26 | 2022-10-25 | 南昌理工学院 | Epidemic disease prediction method based on deep embedded clustering element learning |
| CN115577746A (en) * | 2022-09-15 | 2023-01-06 | 南京辰光融信技术有限公司 | Network intrusion detection method based on meta-learning |
| CN115276006A (en) * | 2022-09-26 | 2022-11-01 | 江苏永鼎股份有限公司 | Load prediction method and system for power integration system |
Non-Patent Citations (1)
| Title |
|---|
| "基于PCC-ML深度学习的微电网多目标协调优化运行";李建杰 等;《电气传动》;第第53卷卷(第第5期期);17-24 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116011657A (en) | 2023-04-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Sun et al. | Using Bayesian deep learning to capture uncertainty for residential net load forecasting | |
| CN111563610B (en) | Building electric load comprehensive prediction method and system based on LSTM neural network | |
| Wu et al. | Evolving RBF neural networks for rainfall prediction using hybrid particle swarm optimization and genetic algorithm | |
| Kizielewicz et al. | Comparison of Fuzzy TOPSIS, Fuzzy VIKOR, Fuzzy WASPAS and Fuzzy MMOORA methods in the housing selection problem | |
| Matijaš et al. | Load forecasting using a multivariate meta-learning system | |
| JP5888640B2 (en) | Photovoltaic power generation prediction apparatus, solar power generation prediction method, and solar power generation prediction program | |
| Farah et al. | Bayesian emulation and calibration of a dynamic epidemic model for A/H1N1 influenza | |
| CN112100911B (en) | Solar radiation prediction method based on depth BILSTM | |
| Guo et al. | An integrated chaotic time series prediction model based on efficient extreme learning machine and differential evolution | |
| Liao | Genetic k-means algorithm based RBF network for photovoltaic MPP prediction | |
| Zhang et al. | Predicting real-time locational marginal prices: A gan-based approach | |
| Mori et al. | Hybrid intelligent method of relevant vector machine and regression tree for probabilistic load forecasting | |
| CN116415177A (en) | Classifier parameter identification method based on extreme learning machine | |
| CN115545333A (en) | Method for predicting load curve of multi-load daily-type power distribution network | |
| CN116826699A (en) | Photovoltaic cluster power prediction method, device, equipment and medium | |
| Madhavi et al. | Multivariate deep causal network for time series forecasting in interdependent networks | |
| CN117829322A (en) | Associated data prediction method based on periodic time sequence and multiple dimensions | |
| Mierzwa | Implementation of multivariate statistical analysis for warning forecasting | |
| CN116011657B (en) | Optimization method, device and system for power distribution network load prediction model based on miniature PMU | |
| Mohammed et al. | Deep learning and neural network-based wind speed prediction model | |
| Naoui et al. | Integrating iot devices and deep learning for renewable energy in big data system | |
| Dan et al. | Application of machine learning in forecasting energy usage of building design | |
| Mele et al. | Machine learning platform for profiling and forecasting at microgrid level | |
| Iwamura et al. | Real-time scheduling for holonic manufacturing systems based on estimation of future status | |
| CN116613732A (en) | Multi-element load prediction method and system based on SHAP value selection strategy |
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 |