CN111193254B - Residential daily electricity load prediction method and device - Google Patents
Residential daily electricity load prediction method and device Download PDFInfo
- Publication number
- CN111193254B CN111193254B CN201911243157.6A CN201911243157A CN111193254B CN 111193254 B CN111193254 B CN 111193254B CN 201911243157 A CN201911243157 A CN 201911243157A CN 111193254 B CN111193254 B CN 111193254B
- Authority
- CN
- China
- Prior art keywords
- data
- encoder
- prediction
- feature vector
- model
- 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
Images
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
Landscapes
- Engineering & Computer Science (AREA)
- Power Engineering (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a residential daily electricity load prediction method and equipment, which comprises the steps of firstly extracting user behavior characteristics by adopting a convolution self-encoder to serve as characteristic vectors, then predicting the characteristic vectors of a user on the current day by utilizing historical day characteristic vectors, finally reconstructing the predicted characteristic vectors by using a decoder to obtain a predicted data set, extracting the characteristics of a residential electricity load data set by utilizing the convolution self-encoder to form the characteristic vectors, capturing the behavior characteristics of the user, predicting the daily change of the characteristic vectors instead of predicting each moment, more conforming to the characteristics of electricity consumption behavior habits of a single residence and effectively improving the accuracy of the daily load prediction of the single residence.
Description
Technical Field
The invention relates to the technical field of electric loads, in particular to a residential daily electric load prediction method and device.
Background
The intelligent power utilization is one of important pillars and main links for constructing a strong intelligent power grid, and is a foundation for realizing the strong intelligent power grid. By means of strong power grid and modern management concepts, technologies such as advanced measurement, efficient control, high-speed communication, rapid energy storage and the like are utilized, rapid market response, justice and accuracy in measurement, real-time data acquisition, various charging modes, efficient and convenient service are realized, and a novel power supply and utilization relation of real-time interaction of a power grid and customer energy flow, information flow and business flow is established.
At present, most residential user power load prediction is performed by adopting a traditional statistical method, for example, a power load curve of a certain time period of a specific cell is drawn by counting power load data of the cell in the certain time period, and then the residential user power load of the cell is predicted by adopting a method combining curve fitting and error analysis.
The inventor of the application discovers that the change of the electricity load of the residential user is different from the load of a factory, a school or some aggregation levels when researching the electricity load of the residential user, so that great uncertainty exists, and meanwhile, the regularity is low. The electricity load of the residential user is predicted by adopting a traditional statistical method, and the prediction precision is low.
Disclosure of Invention
In view of the above, the present invention provides a method and an apparatus for predicting residential daily electrical load, so as to improve the accuracy of predicting residential user electrical load.
The invention provides a method for predicting residential daily electric load based on the above objects, which comprises the following steps,
acquiring residential electricity load and natural environment data;
dividing the acquired electricity load data into a plurality of data sets according to time periods;
constructing a convolution self-encoder based on a data set, and encoding the data set by adopting the convolution self-encoder to obtain a characteristic vector of the data set;
constructing a feature vector prediction model based on the feature vector of the data set, and performing prediction processing on the feature vector of the data set by adopting the feature vector prediction model to obtain a predicted feature vector of the data set;
and decoding the predicted characteristic vector of the data set by adopting a convolution self-encoder to obtain a predicted data set.
Optionally, the constructing a convolution self-encoder based on a data set, and performing encoding processing on the data set by using the convolution self-encoder to obtain a feature vector of the data set includes:
randomly combining the data sets to obtain a combined data set, and dividing the combined data set into an encoder training set and an encoder verification set;
establishing a convolutional self-encoder model comprising an encoder and a decoder;
inputting data in the encoder training set into a convolutional self-encoder model, and performing iterative training on the convolutional self-encoder model until convergence; obtaining a trained convolutional self-encoder model;
verifying the trained convolution self-encoder model by adopting data in an encoder verification set;
and coding the data set by adopting a verified and effective convolution self-coder model to obtain the characteristic vector of the data set.
Optionally, in the building of the convolutional auto-encoder model including the encoder and the decoder, the model depth of the convolutional auto-encoder model is 5 layers of the encoder and 5 layers of the decoder, the number of convolutional kernels is 3, and the number of convolutional kernels in each layer is 16; each layer is provided with batch standardized parameters, and the activation function adopts a nonlinear rectification unit; setting a maximum pooling layer after the 2 nd layer and the 4 th layer of the encoder, setting an upsampling layer after the 2 nd layer and the 4 th layer of the decoder, and adopting an equivalent replication method for upsampling; the output of the encoder is a feature vector, of length 24.
Optionally, the constructing a feature vector prediction model based on the feature vector of the data set, and performing prediction processing on the feature vector of the data set by using the feature vector prediction model to obtain the predicted feature vector of the data set includes:
separating the feature vectors of the data set according to different residences to obtain residential feature vectors;
converting the residential feature vector into available prediction data, and dividing the available prediction data into a training set, a verification set and a test set according to the time sequence;
constructing a feature vector prediction model;
training and verifying the feature vector prediction model by respectively adopting a training set and a verification set until the feature vector prediction model is converged to obtain a trained feature vector prediction model;
and inputting the data of the test set into a trained feature vector prediction model to obtain a predicted feature vector of the data set.
Optionally, the converting the house feature vector into the prediction available data includes a feature vector of a data set of the house, feature vectors of consecutive N data sets preceding the data set, and natural environment data of the N +1 data sets.
Optionally, the dividing of the prediction available data into the training set, the validation set and the test set is dividing the prediction available data into the training set, the validation set and the test set according to a ratio of 08:0.1: 0.1.
Optionally, the decoding, by using a convolutional self-encoder, the prediction feature vector of the data set to obtain a prediction data set, further includes: and carrying out error analysis on the daily power load prediction curve and the real daily power load curve by adopting the average arc tangent absolute value percentage error, wherein the average arc tangent absolute value percentage error meets the following calculation formula:
the method comprises the steps of obtaining a maximum value of a power load of the intelligent ammeter, and obtaining a mean arctangent absolute value percentage error of the power load, wherein MAAPE is a percentage error of the mean arctangent absolute value, N is the number of characteristic vectors in a test set, T is the number of times of collection of the intelligent ammeter in one day, A is a real value of the power load at one moment, and F is a predicted value of the load at one moment.
Optionally, the acquiring residential electricity load and natural environment data includes,
acquiring the power load data of a plurality of residences for years and corresponding local maximum temperature and minimum temperature data every day;
and carrying out processing of eliminating abnormity, deleting redundancy and filling missing on the acquired power load data.
An apparatus for predicting residential daily electrical load, comprising,
an acquisition module: the system is used for acquiring residential electricity load and natural environment data;
the processing module is used for processing the acquired power load data and dividing the processed power load data into a plurality of data sets;
constructing a module: the method is used for constructing a convolution self-encoder based on a data set and constructing a characteristic vector prediction model based on a characteristic vector of the data set;
the coding module: the device is used for encoding the data set to obtain a characteristic vector of the data set;
a prediction module: the characteristic vector prediction method comprises the steps of performing prediction processing on a characteristic vector of a data set to obtain a characteristic vector predicted by the data set;
a decoding module: the characteristic vector used for predicting the data set is decoded to obtain a predicted data set;
an analysis module: for error analysis between the prediction data set and the real data.
From the above, according to the residential daily electricity load prediction method provided by the invention, firstly, the convolutional self-encoder is adopted to extract the user behavior characteristics as the characteristic vector, then the historical day characteristic vector is used to predict the characteristic vector of the user on the same day, finally, the decoder is used to reconstruct the predicted characteristic vector to finally obtain the predicted data set, the characteristic of the residential electricity load data set is extracted by the convolutional self-encoder to form the characteristic vector, the behavior characteristics of the user are captured, the daily change of the characteristic vector is predicted instead of predicting each moment, the characteristic of the electricity utilization behavior habit of a single residence is better met, and the accuracy of the daily load prediction of the single residence is effectively improved.
Meanwhile, through the analysis of the percentage error of the average arctangent absolute value, the prediction error of the method is reduced by 28.6 percent compared with the traditional prediction error, the method effectively reduces the error of the daily load prediction of the house, and improves the prediction precision.
Drawings
FIG. 1 is a flow chart of a residential daily electrical load prediction method according to an embodiment of the present invention;
fig. 2 is a block diagram of the internal structure of the residential daily electrical load prediction device according to the embodiment of the present invention.
Detailed Description
In the following description of the embodiments, the detailed description of the present invention, such as the manufacturing processes and the operation and use methods, will be further described in detail to help those skilled in the art to more fully, accurately and deeply understand the inventive concept and technical solutions of the present invention.
It can be known that, in the prior art, the electricity load curve of a certain time period of a specific cell is drawn by counting the electricity load data of the cell, and then the electricity load of residential users of the cell is predicted by adopting a method of combining curve fitting and error analysis. The prediction method has low prediction accuracy due to the fact that uncertainty of changes of electricity utilization loads of residential users is ignored.
In order to solve the problem of low prediction accuracy, the inventor of the present application has studied and found that the behavior characteristics of the residential users are key factors for providing the residential user electricity load prediction accuracy of the residential district, so the present application provides a residential daily electricity load prediction method, comprising,
acquiring residential electricity load and natural environment data;
dividing the acquired electricity load data into a plurality of data sets according to time periods;
constructing a convolution self-encoder based on a data set, and encoding the data set by adopting the convolution self-encoder to obtain a characteristic vector of the data set;
constructing a feature vector prediction model based on the feature vector of the data set, and performing prediction processing on the feature vector of the data set by adopting the feature vector prediction model to obtain a predicted feature vector of the data set;
and decoding the predicted characteristic vector of the data set by adopting a convolution self-encoder to obtain a predicted data set.
The method comprises the steps of firstly extracting user behavior characteristics by adopting a convolution self-encoder to serve as characteristic vectors, then predicting the characteristic vectors of a user on the same day by utilizing the characteristic vectors of historical days, finally reconstructing the predicted characteristic vectors by using a decoder to obtain a predicted data set, extracting the characteristics of a residential electricity load data set by utilizing the convolution self-encoder to form the characteristic vectors, capturing the behavior characteristics of the user, predicting the daily change of the characteristic vectors instead of predicting each moment, better conforming to the characteristics of electricity consumption behavior habits of a single residence, and effectively improving the accuracy of the daily load prediction of the single residence.
The flow of the residential daily electric load prediction method provided by the embodiment of the application is shown in fig. 1, and the method comprises the following steps.
Step 101: and acquiring residential electricity load and natural environment data.
In one embodiment, step 101 may include obtaining power load data for a plurality of residences over a number of years and corresponding local maximum and minimum temperature data for each day; and carrying out processing of eliminating abnormity, deleting redundancy and filling missing on the acquired power load data.
For example,
(1) the method comprises the steps that for residential users in a certain area, data of the intelligent electric meters of 800 residences in 5 years are obtained, the data are obtained once every 15 minutes, so that 96-point load data can be generated in one residence every day, and meanwhile, data of the highest temperature and the lowest temperature of the local day in the 5 years are obtained;
(2) and (3) executing bad data detection, firstly calculating the average value and the standard deviation of the historical load of a user, and defining the data of which the user load value is more than the average value plus 5 times of the standard deviation or less than 0 as bad data. The detection result shows that the data of nearly 50 users has abnormity, and the abnormal data is removed, so 750 users of available data are left;
(3) executing redundant data deletion, and deleting the exceeding part of the data of which the user day data points exceed 96 points;
(4) and performing missing data filling, and filling data with data points less than 96 points per day of the user in an interpolation mode.
After preprocessing, the retained data are the data required by subsequent prediction.
Step 102: the acquired electricity load data is divided into a plurality of data sets by time period.
In one embodiment, step 102 may include dividing the preprocessed historical electrical load data of each residence into daily electrical load points by day, and forming a daily electrical load curve data set by the daily electrical load points, wherein the daily electrical load curve data set is 1826 days in 5 years, and the daily electrical load curve comprises 96 points, and is a daily electrical load curve of 750 residences.
Step 103: and constructing a convolution self-encoder based on the data set, and encoding the data set by adopting the convolution self-encoder to obtain the characteristic vector of the data set.
In one embodiment, step 103 may include randomly combining the data sets to obtain a combined data set, dividing the combined data set into an encoder training set and an encoder verification set;
establishing a convolutional self-encoder model comprising an encoder and a decoder;
inputting data in the encoder training set into a convolutional self-encoder model, and performing iterative training on the convolutional self-encoder model until convergence; obtaining a trained convolutional self-encoder model;
verifying the trained convolution self-encoder model by adopting data in an encoder verification set;
and coding the data set by adopting a verified and effective convolution self-coder model to obtain the characteristic vector of the data set.
For example,
(1) randomly combining daily electricity load curves of 750 residences in all the data sets segmented in the step 102 to obtain a combined data set, and then segmenting the combined data set into an encoder training set and an encoder verification set according to a random sequence, wherein the corresponding proportion is 90% and 10%;
(2) the self-convolution encoder model comprises an encoder and a decoder, the depth of the model is 5 layers of the encoder and 5 layers of the decoder, in order to prevent gradient disappearance, batch standardization is set for each layer, an activation function adopts a nonlinear rectification unit, and the convolution kernel is 3; the number of convolution kernels of each layer is 16, a maximum pooling layer is arranged behind the 2 nd layer and the 4 th layer of the encoder part, an upsampling layer is arranged behind the 2 nd layer and the 4 th layer of the corresponding decoder part, and the upsampling adopts an equivalent replication method; the output of the encoder is a feature vector with a length of 24;
(3) inputting the daily electric load curves in the encoder training set into a convolution self-encoder model according to 128 loads in each batch, training the convolution self-encoder by adopting an early stopping strategy, stopping training if the reconstruction error of the convolution self-encoder is not reduced within 100 batches, and storing the structure and parameters of the convolution self-encoder; obtaining a trained convolutional self-encoder model;
(4) setting an overfitting threshold as that the verification error is 2 times of the training error, inputting the daily electric load curve in the verification set of the encoder into a training convolution self-encoder for verification, wherein the verification result of the verification set of the encoder shows that the verification error is 1.5 times of the training error and is within the range set by the overfitting threshold, so that the trained convolution self-encoder model is effective;
(5) the entire data set segmented in step 102 is encoded by using a validated convolutional auto-encoder model, and the encoder of the convolutional auto-encoder model outputs the feature vector of each user per day, so that a total of 750 × 1826 and 1369500 feature vectors is generated.
The extraction of the characteristic vector of the electricity utilization behavior of the user is completed.
Step 104: and constructing a feature vector prediction model based on the feature vector of the data set, and performing prediction processing on the feature vector of the data set by adopting the feature vector prediction model to obtain the predicted feature vector of the data set.
In one embodiment, step 104 may comprise: separating the feature vectors of the data set according to different residences to obtain residential feature vectors;
converting the residential feature vector into available prediction data, and dividing the available prediction data into a training set, a verification set and a test set according to the time sequence; converting the house feature vector into prediction available data, wherein the prediction available data comprises the feature vector of one data set of the house, the feature vectors of N continuous data sets before the data set, and the natural environment data of the N +1 data sets; the method comprises the steps that the available prediction data are divided into a training set, a verification set and a test set according to the proportion of 08:0.1: 0.1;
constructing a feature vector prediction model;
training and verifying the feature vector prediction model by respectively adopting a training set and a verification set until the feature vector prediction model is converged to obtain a trained feature vector prediction model;
and inputting the data of the test set into a trained feature vector prediction model to obtain a predicted feature vector of the data set.
For example,
(1) all the feature vectors generated in the step 103 are separated according to different residences, and each user has 1826 day feature vectors, namely the residence feature vectors;
(2) the feature vector of one day of a house is used as a predicted vector, the feature vector of 15 continuous days before the day of the house is used as input data of a feature vector prediction model, the highest temperature and the lowest temperature of the 16 days are added simultaneously to generate a piece of available prediction data, the feature vectors of all days of a user are converted into the available prediction data by adopting the method, and the available prediction data are divided into a training set, a verification set and a test set according to the chronological order, wherein the corresponding proportion is 80%, 10% and 10%.
(3) The long-short term memory network (LSTM) is used to build the prediction model, and the LSTM introduces various gate structures in the cyclic unit, and is calculated as follows:
it＝σ(Wi·[ht-1,xt]+bi)
ft＝σ(Wf·[ht-1,xt]+bf)
ot＝σ(Wo·[ht-1,xt]+bo)
ht＝ot*tanh(Ct)
where i ist、ftAnd otRespectively an input gate, a forgetting gate and an output gate. CtIt is the state of the cell that is,
representing candidate cell states, itDecide whether to update it to cell state CtIn (1). The cell states help the model to better convey gradient information. W and b are parameters and offsets in the cell. Tong (Chinese character of 'tong')Over-conversion input sequence x1,x2,…,xnInputting the data into LSTM to obtain a corresponding hidden state { h }1,h2,…,hn}. These hidden states are used as a signature representation of the sequence for generating the output and at the same time as the input for the next cycle.
The settings were as follows: the unit depth is 3, the number of hidden layer neurons is 64, the forgetting gate bias is initialized to 1, and other parameters are initialized randomly according to (0, 0.1) normal distribution. The input at each time is the feature vector of the day, the output is the predicted value of the feature vector of the day, and the input batch number is 16. The loss function employs a mean square error.
(4) And inputting the data in the training set into the prediction model for training, verifying by using the verification set, and stopping training if the model is converged.
Adopting ADAM optimization algorithm and learning rate attenuation to accelerate convergence: the ADAM algorithm automatically adjusts the learning rate to accelerate learning; if the error of the new verification set does not decrease in 50 training batches, the learning rate is attenuated to 0.1 of the current value; if no new error drop of the verification set occurs within 100 batches, the LSTM model is determined to be converged, and the training is stopped.
(5) The data in the test set is input into a trained feature vector prediction model to generate predicted daily feature vectors, where each user gets 182 days' predicted feature vectors.
Step 105: decoding the predicted characteristic vector of the data set by adopting a convolution self-encoder to obtain a predicted data set, wherein the decoding comprises the following steps: and inputting the daily feature vector obtained by prediction into a decoder of a trained convolution self-encoder, and obtaining a prediction data set, namely a daily electric load prediction curve, by decoding.
In order to verify the error between the residential daily electrical load prediction curve provided by the embodiment of the invention and the current daily load curve of the real situation, the average arctan absolute value percentage error (MAAPE) is adopted for error analysis, the threshold value of the MAAPE is fixed at 0.5, and the average arctan absolute value percentage error meets the following calculation formula:
the method comprises the steps of collecting characteristic vectors of a smart meter, wherein N is the number of the characteristic vectors in a test set, T is the number of times of collection of the smart meter in one day, A is a real value of a moment power load, and F is a predicted value of the moment load.
The average MAAPE of multiple users obtained by the prediction method provided by the embodiment of the invention is 0.34 and is within the threshold value, so the steps of the implementation are successfully completed. Compared with MAAPE of the traditional ARIMA algorithm and MAAPE of the traditional SVR algorithm which are respectively 0.51 and 0.46, the MAAPE obtained by the embodiment is respectively reduced by 32.7 percent and 26.4 percent, so that the method disclosed by the embodiment of the invention effectively reduces the error of daily load prediction of a single house and improves the prediction accuracy.
Based on the above method, the internal structural block diagram of the device for predicting residential daily electrical load provided by the embodiment of the present invention is shown in fig. 2: the device comprises an acquisition module 201, a processing module 202, a construction module 203, an encoding module 204, a prediction module 205, a decoding module 206 and an analysis module 207.
The acquisition module 201 is used for acquiring residential electricity load and natural environment data, and the specific acquisition module 201 acquires data of the smart meters of 800 residences in 5 years, wherein the data is acquired every 15 minutes, and meanwhile, the data of the highest temperature and the lowest temperature of the local day in the 5 years are acquired.
The processing module 202 is configured to process the acquired power load data, and divide the processed power load data into a plurality of data sets; the specific processing module 202 eliminates the abnormal data, and deletes the exceeding part of the data of which the user day data points exceed 96 points; and filling data of less than 96 user day data points by adopting an interpolation mode.
The system comprises a construction module 203, a convolution self-encoder and a feature vector prediction model, wherein the construction module 203 is used for constructing a convolution self-encoder based on a data set and constructing a feature vector prediction model based on a data set feature vector, the construction module 203 specifically constructs convolution self-encoder models of an encoder and a decoder, the model depth is 5 layers of the encoder and 5 layers of the decoder, batch standardization is set for each layer in order to prevent gradient disappearance, a nonlinear rectification unit is adopted as an activation function, and the convolution kernel is 3; the number of convolution kernels of each layer is 16, a maximum pooling layer is arranged behind the 2 nd layer and the 4 th layer of the encoder part, an upsampling layer is arranged behind the 2 nd layer and the 4 th layer of the corresponding decoder part, and the upsampling adopts an equivalent replication method; the output of the encoder is a feature vector, of length 24.
The encoding module 204 is configured to encode the data set to obtain a feature vector of the data set, the specific encoding module 204 encodes all the data sets segmented in step 102 by using a validated convolutional self-encoder model, and an encoder of the convolutional self-encoder model outputs the feature vector of each user every day.
The prediction module 205 is configured to perform prediction processing on the feature vector of the data set to obtain a feature vector predicted by the data set; the specific prediction module 205 is configured to separate feature vectors of the data set according to different residences to obtain residence feature vectors;
converting the residential feature vector into available prediction data, and dividing the available prediction data into a training set, a verification set and a test set according to the time sequence; converting the house feature vector into prediction available data, wherein the prediction available data comprises the feature vector of one data set of the house, the feature vectors of N continuous data sets before the data set, and the natural environment data of the N +1 data sets; the method comprises the steps that the available prediction data are divided into a training set, a verification set and a test set according to the proportion of 08:0.1: 0.1;
constructing a feature vector prediction model;
and respectively adopting a training set and a verification set to train and verify the feature vector prediction model until the feature vector prediction model is converged to obtain the trained feature vector prediction model.
A decoding module 206, configured to perform decoding processing on the predicted feature vector of the data set to obtain a predicted data set; the specific decoding module 206 is configured to decode the predicted feature vector of the data set by using a convolutional self-encoder to obtain a predicted data set.
The analysis module 207 is configured to perform error analysis on the prediction data set and the real data, and the specific analysis module 207 is configured to perform error analysis on the daily power load prediction curve and the real daily power load curve by using the average arctangent absolute value percentage error.
Those of ordinary skill in the art will understand that: the discussion of any embodiment above is meant to be exemplary only, and is not intended to intimate that the scope of the disclosure, including the claims, is limited to these examples; within the idea of the invention, also features in the above embodiments or in different embodiments may be combined, steps may be implemented in any order, and there are many other variations of the different aspects of the invention as described above, which are not provided in detail for the sake of brevity.
The embodiments of the invention are intended to embrace all such alternatives, modifications and variances that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, substitutions, improvements and the like that may be made without departing from the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (6)
1. A residential daily electricity load prediction method is characterized by comprising the following steps,
acquiring residential electricity load and natural environment data;
dividing the acquired electricity load data into a plurality of data sets according to time periods;
randomly combining the multiple data sets to obtain a combined data set, and dividing the combined data set into an encoder training set and an encoder verification set;
establishing a convolutional self-encoder model comprising an encoder and a decoder;
the model depth of the convolutional self-encoder model is 5 layers of an encoder and 5 layers of a decoder, the convolutional kernel is 3, and the number of the convolutional kernels of each layer is 16; each layer is provided with batch standardized parameters, and the activation function adopts a nonlinear rectification unit; setting a maximum pooling layer after the 2 nd layer and the 4 th layer of the encoder, setting an upsampling layer after the 2 nd layer and the 4 th layer of the decoder, and adopting an equivalent replication method for upsampling; the output of the encoder is a feature vector with a length of 24;
inputting data in the encoder training set into a convolutional self-encoder model, and performing iterative training on the convolutional self-encoder model until convergence; obtaining a trained convolutional self-encoder model;
verifying the trained convolution self-encoder model by adopting data in an encoder verification set;
coding the multiple data sets by adopting a verified and effective convolution self-coder model to obtain characteristic vectors of the multiple data sets;
constructing a feature vector prediction model based on the feature vectors of the data sets, and performing prediction processing on the feature vectors of the data sets by adopting the feature vector prediction model to obtain predicted feature vectors of the data sets;
and decoding the prediction characteristic vectors of the plurality of data sets by adopting a convolution self-encoder to obtain a prediction data set.
2. The prediction method according to claim 1, wherein the constructing a feature vector prediction model based on the feature vectors of the plurality of data sets, and performing prediction processing on the feature vectors of the plurality of data sets by using the feature vector prediction model to obtain the predicted feature vectors of the plurality of data sets comprises:
separating the feature vectors of the data sets according to different residences to obtain residential feature vectors;
converting the residential feature vector into available prediction data, and dividing the available prediction data into a training set, a verification set and a test set according to the time sequence;
constructing a feature vector prediction model;
training and verifying the feature vector prediction model by respectively adopting a training set and a verification set until the feature vector prediction model is converged to obtain a trained feature vector prediction model;
and inputting the data of the test set into a trained feature vector prediction model to obtain the predicted feature vectors of the multiple data sets.
3. The prediction method according to claim 2, wherein the converting the house feature vector into the prediction available data comprises feature vectors of one data set of the house, feature vectors of M consecutive data sets preceding the one data set, and natural environment data of the M +1 data sets.
4. The prediction method according to claim 2, wherein the dividing of the prediction available data into the training set, the validation set and the test set is performed by dividing the prediction available data into the training set, the validation set and the test set according to a ratio of 08:0.1: 0.1.
5. The prediction method according to claim 1, wherein after the decoding processing is performed on the predicted feature vectors of the plurality of data sets by using a convolutional auto-encoder to obtain the predicted data sets, the method further comprises: and carrying out error analysis on the daily power load prediction curve and the real daily power load curve by adopting the average arc tangent absolute value percentage error, wherein the average arc tangent absolute value percentage error meets the following calculation formula:
the method comprises the steps of obtaining a maximum quantity of characteristic vectors in a test set, obtaining a maximum quantity of characteristic vectors in the test set, obtaining a real value of a current power load, and obtaining a predicted value of the current load.
6. The prediction method according to claim 1, wherein the obtaining residential electrical load and natural environment data comprises,
acquiring the power load data of a plurality of residences for years and corresponding local maximum temperature and minimum temperature data every day;
and carrying out processing of eliminating abnormity, deleting redundancy and filling missing on the acquired power load data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911243157.6A CN111193254B (en) | 2019-12-06 | 2019-12-06 | Residential daily electricity load prediction method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911243157.6A CN111193254B (en) | 2019-12-06 | 2019-12-06 | Residential daily electricity load prediction method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111193254A CN111193254A (en) | 2020-05-22 |
| CN111193254B true CN111193254B (en) | 2021-10-29 |
Family
ID=70709173
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911243157.6A Active CN111193254B (en) | 2019-12-06 | 2019-12-06 | Residential daily electricity load prediction method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111193254B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112966868A (en) * | 2021-03-13 | 2021-06-15 | 山东大学 | Building load day-ahead prediction method and system |
| CN114444551B (en) * | 2022-04-02 | 2022-06-10 | 西南交通大学 | Aerodynamic load identification method based on wavelet transform and convolution self-encoder |
| CN118691125A (en) * | 2024-08-26 | 2024-09-24 | 深圳江行联加智能科技有限公司 | Power load prediction method, device, equipment and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108197657A (en) * | 2018-01-04 | 2018-06-22 | 成都寻道科技有限公司 | A kind of student's economic situation Forecasting Methodology based on campus data |
| CN108364087A (en) * | 2018-01-11 | 2018-08-03 | 安徽优思天成智能科技有限公司 | A kind of spatio-temporal prediction method of urban mobile tail gas concentration |
| CN108694484A (en) * | 2018-08-30 | 2018-10-23 | 广东工业大学 | A kind of photovoltaic power generation power prediction method |
| CN108921341A (en) * | 2018-06-26 | 2018-11-30 | 国网山东省电力公司电力科学研究院 | A kind of steam power plant's short term thermal load forecasting method encoded certainly based on gate |
| CN110021165A (en) * | 2019-03-18 | 2019-07-16 | 浙江工业大学 | A kind of traffic flow forecasting method based on Autoencoder-LSTM Fusion Model |
| CN110516833A (en) * | 2019-07-03 | 2019-11-29 | 浙江工业大学 | A method of the Bi-LSTM based on feature extraction predicts road traffic state |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10366491B2 (en) * | 2017-03-08 | 2019-07-30 | Siemens Healthcare Gmbh | Deep image-to-image recurrent network with shape basis for automatic vertebra labeling in large-scale 3D CT volumes |
| US10290107B1 (en) * | 2017-06-19 | 2019-05-14 | Cadence Design Systems, Inc. | Transform domain regression convolutional neural network for image segmentation |
| CN108062753B (en) * | 2017-12-29 | 2020-04-17 | 重庆理工大学 | Unsupervised domain self-adaptive brain tumor semantic segmentation method based on deep counterstudy |
| CN108875909A (en) * | 2018-01-22 | 2018-11-23 | 深圳市铱硙医疗科技有限公司 | The classification method and device of medical image |
| CN108764241A (en) * | 2018-04-20 | 2018-11-06 | 平安科技（深圳）有限公司 | Divide method, apparatus, computer equipment and the storage medium of near end of thighbone |
| CN109745062B (en) * | 2019-01-30 | 2020-01-10 | 腾讯科技（深圳）有限公司 | CT image generation method, device, equipment and storage medium |
-
2019
- 2019-12-06 CN CN201911243157.6A patent/CN111193254B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108197657A (en) * | 2018-01-04 | 2018-06-22 | 成都寻道科技有限公司 | A kind of student's economic situation Forecasting Methodology based on campus data |
| CN108364087A (en) * | 2018-01-11 | 2018-08-03 | 安徽优思天成智能科技有限公司 | A kind of spatio-temporal prediction method of urban mobile tail gas concentration |
| CN108921341A (en) * | 2018-06-26 | 2018-11-30 | 国网山东省电力公司电力科学研究院 | A kind of steam power plant's short term thermal load forecasting method encoded certainly based on gate |
| CN108694484A (en) * | 2018-08-30 | 2018-10-23 | 广东工业大学 | A kind of photovoltaic power generation power prediction method |
| CN110021165A (en) * | 2019-03-18 | 2019-07-16 | 浙江工业大学 | A kind of traffic flow forecasting method based on Autoencoder-LSTM Fusion Model |
| CN110516833A (en) * | 2019-07-03 | 2019-11-29 | 浙江工业大学 | A method of the Bi-LSTM based on feature extraction predicts road traffic state |
Non-Patent Citations (6)
| Title |
|---|
| Energy Disaggregation Based Deep Learning Techniques: A pre-processing Stage to Enhance The Household Load Forecasting;Ahmed F. Ebrahim;《IEEE》;20181230;全文 * |
| 基于模糊信息粒化和长短期记忆网络的短期风速预测;殷豪;《电测与仪表》;20190610;全文 * |
| 基于深度学习的电网短期负荷预测方法研究;吴润泽;《现代电力》;20180430;全文 * |
| 朱瑞金.考虑天然气和电负荷之间相关性的短期电负荷预测.《电力系统及其自动化学报》.2019, * |
| 考虑天然气和电负荷之间相关性的短期电负荷预测;朱瑞金;《电力系统及其自动化学报》;20190830;第3-5页；表1；附图6 * |
| 自编码器与PSOA－CNN 结合的短期负荷预测模型;王文卿;《山东大学学报》;20190730;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111193254A (en) | 2020-05-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111193254B (en) | Residential daily electricity load prediction method and device | |
| Ryu et al. | Denoising autoencoder-based missing value imputation for smart meters | |
| Alencar et al. | Hybrid approach combining SARIMA and neural networks for multi-step ahead wind speed forecasting in Brazil | |
| CN105354646B (en) | Power load forecasting method for hybrid particle swarm optimization and extreme learning machine | |
| CN113141008B (en) | Data-driven power distribution network distributed new energy consumption capability assessment method | |
| CN111710150A (en) | Abnormal electricity consumption data detection method based on countermeasure self-coding network | |
| CN111242351A (en) | Tropical cyclone track prediction method based on self-encoder and GRU neural network | |
| CN102495919A (en) | Extraction method for influence factors of carbon exchange of ecosystem and system | |
| CN112330951A (en) | Method for realizing road network traffic data restoration based on generation of countermeasure network | |
| CN109636054A (en) | Solar energy power generating amount prediction technique based on classification and error combination prediction | |
| CN114357670A (en) | Power distribution network power consumption data abnormity early warning method based on BLS and self-encoder | |
| Kosana et al. | A novel hybrid framework for wind speed forecasting using autoencoder‐based convolutional long short‐term memory network | |
| CN117093830A (en) | User load data restoration method considering local and global | |
| CN116384572A (en) | Sequence-to-sequence power load prediction method based on multidimensional gating circulating unit | |
| CN118195353A (en) | Power system load prediction method, device, computer equipment and storage medium | |
| Kavakci et al. | Time Series Prediction of Solar Power Generation Using Trend Decomposition | |
| CN117972579A (en) | Method, device, equipment and storage medium for detecting electricity stealing behavior | |
| CN116227738B (en) | Method and system for predicting traffic interval of power grid customer service | |
| CN117034780A (en) | Multi-scale sub-season precipitation prediction method based on deep learning | |
| CN116861185A (en) | Resident short-term power load prediction system and method based on DCNN-AM-LSTM-AE | |
| CN116050579A (en) | Building energy consumption prediction method and system based on depth feature fusion network | |
| Xie et al. | Deep learning with multisource data fusion in electricity internet of things for electricity price forecast | |
| Huang et al. | Scenario Reduction of Power Systems with Renewable Generations Using Improved Time-GAN | |
| Liu et al. | A novel photovoltaic power output forecasting method based on weather type clustering and wavelet support vector machines regression | |
| Zhao et al. | Ultra-short-term power forecasting of wind farm cluster based on spatio-temporal graph neural network pattern prediction |
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 |