CN115238951A - Power load prediction method and device - Google Patents

Power load prediction method and device Download PDF

Info

Publication number
CN115238951A
CN115238951A CN202210644040.4A CN202210644040A CN115238951A CN 115238951 A CN115238951 A CN 115238951A CN 202210644040 A CN202210644040 A CN 202210644040A CN 115238951 A CN115238951 A CN 115238951A
Authority
CN
China
Prior art keywords
data
neural network
training
lstm
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.)
Pending
Application number
CN202210644040.4A
Other languages
Chinese (zh)
Inventor
刘磊
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiangsu Zhengfang Transportation Technology Co ltd
Original Assignee
Nanjing Wuxiang Cloud Intelligent Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing Wuxiang Cloud Intelligent Technology Co ltd filed Critical Nanjing Wuxiang Cloud Intelligent Technology Co ltd
Priority to CN202210644040.4A priority Critical patent/CN115238951A/en
Publication of CN115238951A publication Critical patent/CN115238951A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Economics (AREA)
  • Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Tourism & Hospitality (AREA)
  • Software Systems (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biophysics (AREA)
  • Power Engineering (AREA)
  • Primary Health Care (AREA)
  • Development Economics (AREA)
  • Water Supply & Treatment (AREA)
  • Quality & Reliability (AREA)
  • Public Health (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a prediction method and a prediction device of a power load, wherein the prediction method comprises the following steps: acquiring user power load data through a power control device; preprocessing the power load data and eliminating error data; performing statistical calculation on the acquired data according to a preset mathematical statistical strategy to obtain N multidimensional statistical characteristic parameters; inputting the data into a CNN deep learning neural network after the processed characteristic data are processed; inputting the sequence with time dependency obtained by CNN into LSTM deep learning neural network according to time sequence, and performing data training; predicting the power load by utilizing the deep learning neural network model; the load prediction method provided by the invention is based on the CNN-LSTM model prediction method, the number of parameters is reduced, the training difficulty is reduced, the problems of training gradient explosion and the like are avoided, and the output prediction result is more accurate than that of the single CNN neural network.

Description

Power load prediction method and device
Technical Field
The invention relates to the field of residential electricity load prediction, in particular to a residential electricity load prediction method and device based on a CNN-LSTM neural network model.
Background
With the rise of new energy sources, the demand of an electric power system is increased, and the importance of electric power load prediction is highlighted. The power load prediction is the basis of the planning work of a power system and is a necessary premise for reasonably arranging power generation, transmission and distribution. Therefore, it is very important to improve the accuracy of the power load prediction.
There are many methods for predicting the power load, but as the speed of power utilization intellectualization is increased, the traditional autonomous regression model cannot be adapted to the environment; due to problems of noise, uncertainty, sensor failure and the like, actual operation data is easily interfered by abnormal values and missing values. Many times, power load prediction needs to be analyzed based on a multivariate time sequence, and the prediction of the power load by using statistical methods such as a traditional autoregressive moving average model (ARMA) and an autoregressive integrated moving average model (ARIMA) is increasingly unable to meet the requirements of practical application. Various machine learning models such as KNN, random forests and support vector machines have good prediction performance under the condition of small data sample size, however, the traditional machine learning model can only extract shallow features and usually needs complex feature engineering. In recent years, many scholars have applied the CNN-LSTM algorithm to the time series prediction problem. The CNN takes the output of the previous hidden layer time step as the input of the current time step and affects the output of the current time step. The unique circulation mechanism of the LSTM is adopted, and good effect is achieved.
Disclosure of Invention
Aiming at the problems, the invention provides a power load prediction method and a power load prediction device based on a CNN-LSTM neural network model, which have higher prediction accuracy.
The invention adopts the neural network of the CNN-LSTM algorithm, fully utilizes the good feature extraction capability of the CNN algorithm and the excellent capability of the LSTM algorithm for modeling aiming at the time sequence, and effectively improves the accuracy of power load prediction.
In order to achieve the above object, a method of predicting a power load includes:
acquiring power consumption data of a user;
carrying out data preprocessing on the electricity utilization data, and rejecting error data;
inputting the preprocessed data into a CNN deep learning neural network to extract characteristic information, and obtaining a data sequence with time dependence;
inputting the data with the extracted features into an LSTM deep learning neural network for training to obtain a training model;
and predicting the electrical load by using the trained model.
As a preferred technical solution, acquiring the power consumption data of the user further includes: collecting user electricity utilization data of a preset time period every day through an electricity utilization load sensor, collecting the user electricity utilization data into a data summary sheet, and ensuring that the user electricity utilization data is collected every 15 minutes when the data are collected;
as a preferred technical solution, the data preprocessing is performed on the electricity consumption data to remove error data, and the method further includes:
removing obvious error information in the data according to prior experience;
integrating and calculating whether the data have larger difference, if so, rejecting the data, and otherwise, reserving the data;
the data which are basically preprocessed are respectively sorted and stored into different categories according to time, acquisition conditions and other attribute requirements;
and performing mathematical statistics through the time attribute, and storing the data volume of at least 90 days in the data set.
As a preferred technical solution, the data that has been basically preprocessed are respectively sorted and stored into different categories according to time, acquisition conditions, and other attribute requirements, and the method further includes:
generating a data set containing time and acquisition characteristics according to the acquisition time and the acquisition characteristics;
the data set is as follows 8:1:1, carrying out data division on a training set, a test set and a verification set.
As a preferred technical solution, the data from which the features have been extracted is input into an LSTM deep learning neural network for training to obtain a training model, further comprising:
inputting the training set into a CNN neural network;
activating by Relu function, and performing pooling treatment by max-pooling;
dropout with a probability of 0.3 was added to prevent overfitting;
the output is data with temporal feature dependence and serves as the input of the LSTM neural network.
As a preferred technical solution, the method for predicting the electrical load by using the trained model further includes:
inputting the sequences with time dependency extracted according to the CNN into the LSTM in time sequence as the input of each timestamp;
in order to enable the model to have better expandability and accuracy, an Attention mechanism is introduced;
training the LSTM neural network according to a seq2seq mode by utilizing the divided data sets;
and mapping the Attentionvalue to an output value, namely a predicted value of the model, through the last full connection layer to obtain a predicted result.
In another aspect, the present invention provides a power load prediction apparatus including:
the acquisition unit is used for acquiring power consumption data of a user;
the preprocessing unit is used for preprocessing the power utilization data and eliminating error data;
the characteristic extraction unit is used for inputting the preprocessed data into the CNN deep learning neural network to extract characteristic information so as to obtain a data sequence with time dependence;
the training unit is used for inputting the data with the extracted features into an LSTM deep learning neural network for training to obtain a training model;
and the prediction unit is used for predicting the electrical load by using the trained model.
In the above prediction apparatus, preferably, the data from which the features have been extracted is input to an LSTM deep learning neural network for training to obtain a training model, and the prediction apparatus further includes:
inputting the training set into a CNN neural network;
activating by a Relu function, and performing pooling treatment by max-pooling;
dropout with a probability of 0.3 was added to prevent overfitting;
the output is data with temporal feature dependence and serves as the input of the LSTM neural network.
In the above prediction device, it is preferable that the power consumption load prediction is performed using a trained model, and the prediction device further includes:
inputting the sequences with time dependency extracted according to the CNN into the LSTM in time sequence as the input of each timestamp;
in order to enable the model to have better expandability and accuracy, an Attention mechanism is introduced;
training the LSTM neural network according to a seq2seq mode by using the divided data set;
mapping the Attentitionvalue to an output value, namely a predicted value of the model, through the last full connection layer to obtain a prediction result.
Compared with the prior art, the invention has the following beneficial effects:
1) The invention can conveniently adopt the prior public data set and deep learning frame to construct and predict the model; the method selects a personal household electricity data set provided by a UCI machine learning knowledge base, and builds a model by using a deep learning framework Keras and TensorFlow; the cost is low, and the implementation is easy;
2) Aiming at the existing data, a method of error elimination is utilized to eliminate partial error data; aiming at the missing part, a linear interpolation method is used for supplementing, so that the model can be closer to real data, and the validity of the model can be better verified;
3) The method fully utilizes the capability of CNN neural network to extract features and the processing mode and the processing capability of the LSTM neural network for the features of the time sequence, has stronger generalization capability and prediction precision, and has higher robustness.
Drawings
FIG. 1 is a flow chart of a method of predicting a power load according to the present invention;
FIG. 2 is a feature extraction simulation diagram of the CNN deep learning network model provided by the present invention;
FIG. 3 is a training simulation diagram of the LSTM deep learning network model provided by the present invention;
fig. 4 is a block diagram of a power load prediction apparatus according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
As used in this application and the appended claims, the terms "a," "an," "the," and/or "the" are not intended to be inclusive in the singular, but rather are intended to be inclusive in the plural unless the context clearly dictates otherwise. In general, the terms "comprises" and "comprising" merely indicate that steps and elements are included which are explicitly identified, that the steps and elements do not form an exclusive list, and that a method or apparatus may include other steps or elements.
The relative arrangement of the components and steps, the numerical expressions, and numerical values set forth in these embodiments do not limit the scope of the present application unless specifically stated otherwise. Meanwhile, it should be understood that the sizes of the respective portions shown in the drawings are not drawn in an actual proportional relationship for the convenience of description. Techniques, methods, and apparatus known to those of ordinary skill in the relevant art may not be discussed in detail but are intended to be part of the specification where appropriate. In all examples shown and discussed herein, any particular value should be construed as exemplary only and not as limiting. Thus, other examples of the exemplary embodiments may have different values. It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, further discussion thereof is not required in subsequent figures.
Referring to fig. 1, the present invention provides a method for predicting a power load, including the steps of:
s10: acquiring power consumption data of a user;
specifically, acquiring daily electrical load data: respectively acquiring power load data of an industrial park and power load data of daily residents;
in the embodiment, the 2006.12-2010.11 electricity consumption data set of nearly four years of a family of Beijing, china is obtained from the UCI machine learning knowledge base. The data set was a multivariate time series data set, with a total of 2075259 time series power usage collected at a 1 minute sample rate. The data set is composed of 9 attribute information. The date and time are time information, the active power is power consumption data, and the reactive power, the voltage, the current intensity, the kitchen active electric energy, the laundry active electric energy, the electric water heater and the air conditioner active electric energy are characteristic data.
S20: carrying out data preprocessing on the electricity utilization data, and rejecting error data;
specifically, 1) the obviously existing error data are removed by direct and intuitive judgment;
2) Potential error data are eliminated again by using a mathematical statistical method to obtain relatively correct data;
3) Completing by using a linear interpolation method; on the whole, the electricity consumption is greatly influenced by the activities of residents, the fluctuation time and the fluctuation amplitude have strong randomness, but the trend is obvious in a short time; the formula of the linear interpolation is shown as (1);
Figure BDA0003685215920000051
in the formula, y ij : indicating the insertion value of the j characteristic in the ith row; x is the number of 0 : the row number value of a five-data missing row in the previous row closest to the ith row is represented; x is a radical of a fluorine atom n : watch (CN)Showing the row number of the row without data loss in the next day closest to the ith row; x is the number of i : a row number representing an ith row; y is 0j : indicating j as a feature in the x-th 0 The value of the row; y is nj Indicating j as a feature in the x n The value of the row.
4) The data of the last three years are divided into a training set, the data of the last year are divided into a test set and a verification set again, and the secondary division proportion is 3:1 (first 9 months, last 3 months).
S30: inputting the preprocessed data into a CNN deep learning neural network to extract characteristic information, and obtaining a data sequence with time dependence;
specifically, 1) CNN feature extraction is carried out, then Relu function is used for activation, then max-posing is carried out for pooling treatment, and finally dropout with the probability of 0.3 is added to prevent overfitting;
2) Outputting a section of sequence at the end as the input of the LSTM; wherein the CNN network simulation diagram is shown in fig. 2;
s40: inputting the data with the extracted features into an LSTM deep learning neural network for training to obtain a training model;
s40, inputting the data with the extracted features into an LSTM deep learning neural network for training to obtain a training model;
in particular, long-short term memory (LSTM) is a variant of Recurrent Neural Networks (RNN) that captures the temporal dependence of time series and achieves good results in time series prediction;
1) Performing feature extraction through CNN to obtain a time-dependent sequence, and inputting the time-dependent sequence into the LSTM as the input of each time step; wherein the LSTM time series prediction network is shown in fig. 3;
2) Introduction of the Attention mechanism: in the process of the last step, in order to improve the expandability and the accuracy of the model, an Attention mechanism is introduced, the output of each time step of the LSTM, namely oi, is weighted and summed with a layer of weight and output through softmax, a is obtained and is used as an Attention coefficient, and the Attention Value is obtained by weighting with output O; wherein the formula is shown as (2), (3) and (4);
M＝sigmoid(O) (2)
α＝softmax(ω T M) (3)
s＝Oα T (4)
(2) In the formula, M: representing the Attention Value obtained by a sigmoid activation function; o: representing the output value after the CNN network extracts the characteristics;
(3) In the formula, α: representing the output value after passing through the full connection layer for one time; ω: representing the output of each time step of the LSTM; t: a transformation matrix representing usage; m: representing an output value obtained by a sigmoid activation function, and taking the output value as an input value;
(4) In the formula, s: a score vector representing the LSTM output; o: representing the output value after the CNN network extracts the characteristics; t: the transformation matrix used is indicated.
Training a data model: training an LSTM according to a seq2seq mode, using a background-Through-Time (BPTT) to perform reverse transfer of an error, wherein a loss function is a mean square error; wherein the loss function formula is shown as (5);
Figure BDA0003685215920000061
wherein, loss: representing the mean square error; n: represents the total number of samples; yi: actual value representing the ith time; output i : indicates the predicted value at the i-th time.
Model output and data prediction: mapping the Attention value to an output value through a layer of full connection layer, namely a predicted value of the model; the output result is shown in equation (6).
output＝sigmoid(ω T s) (6)
In the formula, output: representing a predicted output value; ω: representing the output value after passing through the full connection layer once; t: representing a transformation matrix; s: a score vector representing the LSTM output.
In the general prediction problem, when a long sequence is faced, processing some dependency relationships in context can get into a bottleneck. One solution is by introducing an Attention mechanism that allows the model to focus on the desired part of the input sequence. The model after introduction of the Attention mechanism can amplify the effect of the relevant part in the input sequence to achieve better effect than the model without Attention. Generally, the expression attention mechanism can be viewed as a weighted sum. Therefore, in order to improve the extensibility and accuracy of the integrated neural network model, an Attention mechanism is introduced. And weighting and summing the output of each time step, namely oi, of the LSTM with a layer of weight and outputting the sum through softmax to obtain a as an Attention coefficient, and weighting the sum with output O to obtain the Attention Value.
3) Training a data model: training the LSTM according to a seq2seq mode, and performing reverse transmission of errors by using a background-Through-Time (BPTT);
4) Model output and data prediction: the Attention value is mapped to an output value through a layer of full connection layer, namely a predicted value of the model.
(5) Verifying accuracy and effectiveness of CNN-LSTM power load prediction model
1) 4 common performance evaluation indicators were used, mean Square Error (MSE), root Mean Square Error (RMSE), mean Absolute Error (MAE), and mean relative error (MAPE).
2) On the basis of sampling the original data set per minute, resampling is carried out according to different time units such as every hour, every day, every week and the like, so as to verify the performance of the CNN-LSTM neural network power load prediction model under different time resolutions.
3) Experiments show that the CNN-LSTM hybrid neural network model used at present has higher robustness.
S50: and predicting the electrical load by using the trained model.
In another embodiment, the present invention further provides a power load prediction apparatus, as shown in fig. 4, wherein the identification apparatus includes:
an obtaining unit 100, configured to obtain power consumption data of a user; it should be noted that, since the specific obtaining manner and process are already described in detail in the step S10 of the method for obtaining the power data, they are not described herein again.
The preprocessing unit 200 is used for preprocessing the power consumption data and eliminating error data; it should be noted that, since the specific preprocessing method and process are already described in detail in step S20 of the power load prediction method, detailed description thereof is omitted here.
The feature extraction unit 300 is configured to input the preprocessed data into a CNN deep learning neural network to extract feature information, so as to obtain a data sequence with time dependency; it should be noted that, since the specific feature extraction manner and process are already described in detail in the step S30 of the power load prediction method, they are not described herein again.
The training unit S40 is used for inputting the data with the extracted features into an LSTM deep learning neural network for training to obtain a training model; it should be noted that, since the specific training method and process are already described in detail in step S40 of the power load prediction method, they are not described herein again.
The prediction unit S50 is configured to perform the electrical load prediction by using the trained model, and it should be noted that, since the specific prediction manner and process are already described in detail in the step S50 of the electrical load prediction method, they are not described herein again.
In addition, an embodiment of the present invention further provides a computer-readable storage medium, where the computer-readable storage medium may store a program, and when the program is executed, the program includes some or all of the steps of any one of the power load prediction methods described in the above method embodiments.
Finally, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated unit, if implemented as a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable memory. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a memory and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned memory comprises: a U-disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a removable hard disk, a magnetic or optical disk, and other various media capable of storing program codes.
Those skilled in the art will appreciate that all or part of the steps in the methods of the above embodiments may be implemented by associated hardware instructed by a program, which may be stored in a computer-readable memory, which may include: flash memory disks, read-only memory, random access memory, magnetic or optical disks, and the like.
It should be noted that the numerous details included in the above description are merely exemplary of the invention and are not limiting of the invention. In other embodiments of the invention, the method may have more, fewer, or different steps, and the order, inclusion, function, etc. of the steps may be different than that described or illustrated.

Claims (10)

1. A method for predicting a power load, comprising:
acquiring power consumption data of a user;
carrying out data preprocessing on the electricity utilization data, and rejecting error data;
inputting the preprocessed data into a CNN deep learning neural network to extract characteristic information, and obtaining a data sequence with time dependence;
inputting the data with the extracted features into an LSTM deep learning neural network for training to obtain a training model;
and predicting the electrical load by using the trained model.
2. The prediction method of claim 1, wherein obtaining electricity consumption data of a user further comprises:
through the power consumption load sensor, collect the user's power consumption data of presetting the period of time every day, add up to the data summary table, and when data acquisition, guarantee to gather once every 15 minutes.
3. The method according to claim 2, wherein the electricity consumption data is subjected to data preprocessing to remove error data, and further comprising:
removing obvious error information in the data according to prior experience;
integrating and calculating whether the data have larger difference, if so, rejecting the data, and otherwise, reserving the data;
the data which are basically preprocessed are respectively sorted and stored into different categories according to time, acquisition conditions and other attribute requirements;
and performing mathematical statistics through the time attribute, and storing the data volume of at least 90 days in the data set.
4. The prediction method according to claim 3, wherein the data that has been basically preprocessed are sorted and stored into different categories according to time, collection conditions and other attribute requirements, and further comprising:
generating a data set containing time and acquisition characteristics according to the acquisition time and the acquisition characteristics;
the data set is as follows 8:1:1, dividing data into a training set, a test set and a verification set.
5. The prediction method of claim 4, wherein the data with extracted features is input into an LSTM deep learning neural network for training, and a training model is obtained, further comprising:
inputting the training set into a CNN neural network;
activating by Relu function, and performing pooling treatment by max-pooling;
dropout with a probability of 0.3 is added to prevent overfitting;
the output is data with temporal feature dependence and serves as the input of the LSTM neural network.
6. The method of predicting according to claim 5, wherein the power load prediction is performed by using the trained model, further comprising:
inputting the sequences with time dependency extracted by the CNN into the LSTM in a time sequence as the input of each timestamp;
in order to enable the model to have better expandability and accuracy, an Attention mechanism is introduced;
training the LSTM neural network according to a seq2seq mode by utilizing the divided data sets;
and mapping the Attentionvalue to an output value, namely a predicted value of the model, through the last full connection layer to obtain a predicted result.
7. An apparatus for predicting an electric load, comprising:
the acquisition unit is used for acquiring power utilization data of a user;
the preprocessing unit is used for preprocessing the power utilization data and eliminating error data;
the characteristic extraction unit is used for inputting the preprocessed data into the CNN deep learning neural network to extract characteristic information so as to obtain a data sequence with time dependence;
the training unit is used for inputting the data with the extracted features into an LSTM deep learning neural network for training to obtain a training model;
and the prediction unit is used for predicting the electrical load by using the trained model.
8. The prediction apparatus according to claim 7, wherein the data from which the features have been extracted is input into an LSTM deep learning neural network for training to obtain a training model, further comprising:
inputting the training set into a CNN neural network;
activating by Relu function, and performing pooling treatment by max-pooling;
dropout with a probability of 0.3 is added to prevent overfitting;
the output is data with temporal feature dependence and serves as the input of the LSTM neural network.
9. The prediction device of claim 8, wherein the trained model is used for power load prediction, further comprising:
inputting the sequences with time dependency extracted according to the CNN into the LSTM in time sequence as the input of each timestamp;
in order to enable the model to have better expandability and accuracy, an Attention mechanism is introduced;
training the LSTM neural network according to a seq2seq mode by utilizing the divided data sets; mapping the Attentitionvalue to an output value, namely a predicted value of the model, through the last full connection layer to obtain a prediction result.
10. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of a method for predicting an electrical load according to any one of claims 1 to 6.
CN202210644040.4A 2022-06-09 2022-06-09 Power load prediction method and device Pending CN115238951A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210644040.4A CN115238951A (en) 2022-06-09 2022-06-09 Power load prediction method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210644040.4A CN115238951A (en) 2022-06-09 2022-06-09 Power load prediction method and device

Publications (1)

Publication Number Publication Date
CN115238951A true CN115238951A (en) 2022-10-25

Family

ID=83668712

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210644040.4A Pending CN115238951A (en) 2022-06-09 2022-06-09 Power load prediction method and device

Country Status (1)

Country Link
CN (1) CN115238951A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117688343A (en) * 2024-02-04 2024-03-12 成都帆点创想科技有限公司 LTV prediction method and system for multi-task learning LSTM-Attention framework

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117688343A (en) * 2024-02-04 2024-03-12 成都帆点创想科技有限公司 LTV prediction method and system for multi-task learning LSTM-Attention framework
CN117688343B (en) * 2024-02-04 2024-05-03 成都帆点创想科技有限公司 LTV prediction method and system for multi-task learning LSTM-Attention framework

Similar Documents

Publication Publication Date Title
CN111754026B (en) Photovoltaic power station group power prediction method and device, computer equipment and storage medium
EP4085387A1 (en) Method and apparatus for predicting power consumption, device and readiable storage medium
CN112381673B (en) Park electricity utilization information analysis method and device based on digital twin
CN110212524A (en) A kind of region Methods of electric load forecasting
Cheng et al. Enhanced state estimation and bad data identification in active power distribution networks using photovoltaic power forecasting
Liu et al. Heating load forecasting for combined heat and power plants via strand-based LSTM
CN109359665B (en) Household appliance load identification method and device based on support vector machine
CN109934303A (en) A kind of non-invasive household electrical appliance load recognition methods, device and storage medium
CN111815060A (en) Short-term load prediction method and device for power utilization area
CN116451117A (en) Power data anomaly detection method based on federal learning
CN106295877B (en) Method for predicting electric energy consumption of smart power grid
CN112418545A (en) Load characteristic and model fusion based electricity sales amount prediction method and system
Su Analysis of optimisation method for online education data mining based on big data assessment technology
CN115374998A (en) Load prediction method based on attention mechanism and CNN-BiGRU
CN115238951A (en) Power load prediction method and device
CN114971090A (en) Electric heating load prediction method, system, equipment and medium
CN116628534A (en) Method for dividing energy dynamic images for park based on electric power big data
CN113762591B (en) Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning
CN114638170A (en) Electric energy metering device fault diagnosis method and system based on data mining technology
CN114358362A (en) Electric vehicle load prediction method under condition of data shortage
CN113722970B (en) Photovoltaic power ultra-short-term online prediction method
CN110852597B (en) Electricity consumption peak period resident load ratio calculation method based on generation of countermeasure network
Li et al. Short-term Power Load Forecasting based on Feature Fusion of Parallel LSTM-CNN
Zhang et al. Load prediction based on depthwise separable convolution model
Ding et al. Short-Term Load Forecasting Reliability in Power Plant of Cyber-Physical Energy System Considering Adaptive Denoising

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
TA01 Transfer of patent application right
TA01 Transfer of patent application right

Effective date of registration: 20230406

Address after: 211100 3286, floor 3, Chuangye building, No. 1009, Tianyuan East Road, Jiangning District, Nanjing, Jiangsu Province (Jiangning high tech Zone)

Applicant after: Jiangsu Zhengfang Transportation Technology Co.,Ltd.

Address before: Room 921, Building 6, No. 6, Sanhong Road, Yuhuatai District, Nanjing, Jiangsu Province, 210012

Applicant before: Nanjing Wuxiang Cloud Intelligent Technology Co.,Ltd.