CN113962364A - Multi-factor power load prediction method based on deep learning - Google Patents

Multi-factor power load prediction method based on deep learning Download PDF

Info

Publication number
CN113962364A
CN113962364A CN202111232185.5A CN202111232185A CN113962364A CN 113962364 A CN113962364 A CN 113962364A CN 202111232185 A CN202111232185 A CN 202111232185A CN 113962364 A CN113962364 A CN 113962364A
Authority
CN
China
Prior art keywords
data
power
lstm
sequence
load data
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.)
Granted
Application number
CN202111232185.5A
Other languages
Chinese (zh)
Other versions
CN113962364B (en
Inventor
朱敏
明章强
闫建荣
张万利
赵志龙
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.)
Sichuan University
Original Assignee
Sichuan University
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 Sichuan University filed Critical Sichuan University
Priority to CN202111232185.5A priority Critical patent/CN113962364B/en
Publication of CN113962364A publication Critical patent/CN113962364A/en
Application granted granted Critical
Publication of CN113962364B publication Critical patent/CN113962364B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • 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
    • 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
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
    • H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
    • 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
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P80/00—Climate change mitigation technologies for sector-wide applications
    • Y02P80/20—Climate change mitigation technologies for sector-wide applications using renewable energy
    • 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)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Power Engineering (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Supply And Distribution Of Alternating Current (AREA)

Abstract

The invention discloses a multi-factor power load forecasting method based on deep learning, which comprises the steps of firstly completing the acquisition and storage of data, including power load data and environmental influence data; preprocessing and standardizing data based on abnormal data detection, autoregressive interpolation and sequence data normalization of a k-proximity algorithm and an improved DBSCAN algorithm; then, an improved CNN-LSTM electrical load prediction model is provided, and a CNN characteristic extraction module is adopted to learn local characteristics of input data; then inputting the data into an LSTM sequence learning model, and extracting sequence characteristic information of input data; meanwhile, introducing a self-attribute mechanism into the LSTM for learning the characteristics of the LSTM hidden layer, and realizing the extraction of key characteristics by distributing different attention weights so as to improve the final prediction precision; and finally, predicting the electrical load. The invention can promote the digital upgrading of the power grid, meet the individual requirements of users, and realize the association analysis of industry, the dispatching of power generation, the prediction of power utilization trend, the guidance of repeated work and production and the like.

Description

Multi-factor power load prediction method based on deep learning
Technical Field
The invention relates to the crossing field of artificial intelligence and power load prediction, in particular to a multi-factor user power load prediction method based on deep learning.
Background
In recent years, with the rapid and high-quality development of national economy, the living standard of people is continuously improved, and the demand for electric power is also continuously increased. Countries also accelerate the deployment of power projects to ensure the sufficiency of power supplies. However, due to the lagging generation link of the related power generation project, a decision maker in the power department cannot accurately grasp the load change of the power grid, so that the problems of decision making errors and the like occur, a reasonable power supply and demand relationship cannot be maintained, and a large amount of power resources are wasted. The following significance is achieved for predicting the user electrical load: (1) the method comprises the steps of mining the power utilization mode of the user, analyzing power utilization information of the user, mining the power utilization rule of the user, and further exploring the reason for the difference between the power utilization mode of the user and the power utilization rule of the user by combining factors such as regions, time and the like. Aiming at the prediction result of the power load of the power grid, the power department can conveniently schedule power resources and provide personalized service among different users in different areas; (2) the geographic scale guides the power transmission, and the prediction of the power load of the power grid on the geographic scale can be combined with the difference of the power generation amount among the regions, so that the reasonable scheduling and transmission of power resources by a power department are facilitated; (3) the method is used for assisting in realizing a visual interaction system, building a prediction and visual analysis model based on power data for a power load prediction result, assisting with a prediction method with a good analysis and comparison effect, and realizing a power grid power load prediction system, so that the predicted power loads of power users in different regions, different time periods and different types can be vividly displayed, and a series of influence factors of the power loads can be reflected in the visual system.
The power grid load prediction system considers influence factors such as users, regions, time, environment, climate and the like, and realizes power load prediction at a certain time or a certain period in the future by constructing a training model. In the previous methods, a prediction model is established by simulating a logistic growth curve model, a multiple linear regression analysis model, a gray prediction model and a neural network model, power supply load and power consumption in a Harbin region are predicted, and a method with higher precision is determined by comparative analysis. Some methods are based on a BP neural network, and the influence of multiple factors such as temperature and the like is considered, so that an output result with higher accuracy is obtained. Researchers put forward a Lasso-PCA data reduction and feature extraction model to reduce the calculated amount and parameters of the model, improve the execution efficiency of the model, innovatively apply an improved adaptive genetic algorithm to optimize a BP neural network training process, and the method has higher prediction precision on the power load through analysis and demonstration. Aiming at the defects of single-scale research, the recurrent neural network prediction model based on the back propagation algorithm of time sequence decomposition can be established by synthesizing the influences of various factors, and can predict the power consumption of a user in the future period. Researchers also put forward a power utilization prediction model based on a water wave optimization algorithm to improve a radial basis function neural network, and the problem that power utilization load prediction is inaccurate under the condition based on a user level is solved. The RBF neural network hidden layer center optimization and the expansion constant parameter optimization further prove that the method has higher accuracy.
With the development of smart power grids and the increasing demand of the power industry, the importance of power load prediction is increasingly shown, and the requirement on the power load prediction precision is higher and higher. Summarizing the current similar research and technical findings: most of the existing power grid load prediction methods are biased to traditional algorithms such as a regression analysis method, a time series method and the like, the methods have defects, influence of complex factors such as meteorological data and the like cannot be considered, and the difference between a prediction result and a true value is large; or the method of machine learning is only oriented to a single data source, such as the current power load and the historical power load, although the short-term prediction effect is improved compared with the traditional method, the user power load is often greatly influenced by environmental factors, such as weather, temperature, humidity, holidays and the like, so that the prediction accuracy of the previous method is poor, and the deviation of the prediction result is large. Therefore, exploring new multi-factor electrical load forecasting based on deep learning has become a current research focus.
Disclosure of Invention
In view of the above problems, the present invention provides a method for predicting electrical load of a multi-factor user based on deep learning, so as to achieve the purpose of accurately predicting electrical load under different user categories, regions and different environmental factors. The technical scheme is as follows:
a multi-factor power load prediction method based on deep learning comprises the following steps:
step 1: acquiring power load data including different areas, years and power utilization categories and environmental influence data including temperature, humidity and wind power, and storing the data in a sqlite database;
step 2: the method for cleaning and preprocessing the power load data and the environmental impact data comprises the following steps: abnormal data detection, autoregressive interpolation and sequence data normalization based on a k-proximity algorithm and an improved DBSCAN algorithm;
and step 3: introducing a self-attention mechanism, performing feature reconstruction on a hidden layer of the LSTM to realize end-to-end deployment, constructing an electrical load data prediction neural network model based on the improved CNN-LSMT, and using the electrical load data and the environmental impact data obtained by processing in the step (2) as a training set and a test set;
and 4, step 4: and carrying out the electric load prediction through the electric load data prediction neural network model of the improved CNN-LSMT.
Further, the electric load data comprises a user number, a region to which the user belongs, an urban/rural network, a user classification, an electricity utilization type, a client type, a voltage grade, an industry type, a contract capacity, an operation capacity, a first power transmission date, a frozen electric quantity date and a frozen electric quantity of 15 minutes; the environmental impact data comprises the current date, the region, the highest temperature, the lowest temperature, the average humidity, the average wind power, the weather type, the current temperature of 15 minutes and whether the current temperature is a holiday or not; and the time of the environmental impact data is synchronized with the time of the electrical load data.
Furthermore, the abnormal data detection based on the k-proximity algorithm and the improved DBSCAN algorithm in the step 2 specifically comprises:
step 2.1: defining the average distance between a sample and an adjacent sample as the abnormal score of the sample, obtaining the abnormal score of the electricity load data at each moment by using a modified KNN algorithm, and taking the total distance from each cluster to each k adjacent samples as the final abnormal score;
load data c of electricity consumption at a certain time iiK adjacent sets N ofk(ci) Expressed as:
Figure BDA0003316420310000021
Figure BDA0003316420310000031
in the formula (I), the compound is shown in the specification,
Figure BDA0003316420310000032
denotes ciOne of the k neighboring points of dk(ci) Is ciThe average distance of k neighbors of (a),
Figure BDA0003316420310000033
denotes ciAnd its neighboring point
Figure BDA0003316420310000034
The distance of (d);
electrical load data ciIs expressed as:
Figure BDA0003316420310000035
in the formula, Nk(ci) Denotes ciSet of k neighboring points
Finally, outputting the first m clusters of the abnormal score ranking list as abnormal values of the electric load data;
step 2.2: adopting an improved cluster-based anomaly detection algorithm DBSCAN, firstly utilizing local parameters to realize density clustering of data of small samples, then carrying out iterative clustering on local clustering results to realize a final global clustering result, and marking points which do not belong to any cluster as anomalous points; the method specifically comprises the following steps:
step a) updating parameters:
setting the size M of a cluster sliding window, calculating the average distance difference of electric load data in the window, setting the former k adjacent electric loads as MinPst, and setting the Euclidean distance between the electric load data as Eps;
setting a weight for each load data to reduce the impact on the final clustering result, weight w (c)i,cj) The calculation formula of (a) is as follows:
Figure BDA0003316420310000036
Figure BDA0003316420310000037
wherein, Cov (c)i,cj) Electrical load data c at time iiAnd the electric load data c at the time of jjCovariance of (a), Var (c)i) Electrical load data c at time iiVariance of (c), Var (c)j) The power load data c at time jjThe variance of (a); r (c)i,cj) Is denoted by ciAnd cjThe smaller the value, the greater the correlation;
step b) obtaining an abnormal detection result of the electric load data by analyzing the clustering result:
in the clustering process, setting the abnormal point marked for the first time as a candidate abnormal point, setting the abnormal score plus 1, entering the next candidate abnormal point in the cyclic iterative clustering process, and updating the abnormal score; if the anomaly score S is equal to the cluster number C, the anomaly point is marked.
Further, the autoregressive interpolation in step 2 has the following steps:
completing the missing electric load data value by adopting a Lagrange interpolation method so that the polynomial y of n-1 is a0+a1x+a2x2+…+an-1xn-1Coordinate (x) passing through n points1,y1),(x2,y2),(x3,y3),…,(xn,yn) Then the functional expression for lagrange interpolation is expressed as:
Figure BDA0003316420310000041
wherein x isiAnd xjI and j time instants, y, representing the power consumers, respectivelyiA power load indicating an i-th time of the power consumer; n represents the total number of times of the electrical load.
Further, the processing formula for the normalization of the sequence data in step 2 is:
Figure BDA0003316420310000042
wherein X ═ { X ═ X1,X2,X3…,XNRepresents the same category of electrical load data or environmental impact data; the normalized value is Y ═ Y1,Y2,Y3…,YN}，l∈[1,N]And N is the total number of certain types of electrical load data or environmental impact data.
Further, the step 3 specifically includes:
step 3.1: loading and preparing for network model data
Loading a data set, dividing a training set, a verification set and a test set, constructing a Dataloader as a data reader for the training set and the test set respectively, calculating data of each batch during model training, loading the data into a memory by the Dataloader according to the size of the batch, and disordering the data of each batch to improve the robustness of the model training;
step 3.2: construction of electrical load data prediction neural network model based on improved CNN-LSMT
The neural network model comprises a CNN feature learning module, an LSTM sequence learning module and an attention mechanism module:
1) the CNN feature learning module comprises three one-dimensional convolution layers, wherein a Max painting layer and a ReLu layer are added between two continuous convolution layers; learning the characteristics of the normalized power load data and the normalized environmental impact data through convolution operation, and using the characteristics as a characteristic diagram output by the convolution layer; adding a MaxPholing layer to relieve the limitation of invariance of the generated feature map, and activating a function ReLu to enhance the capability of the model to learn a complex structure;
2) the LSTM sequence learning module comprises three LSTM layers, each layer comprising twenty neurons; the first two LSTM layers output the complete sequence of the hidden state, and the last LSTM layer outputs the last time step of the hidden state;
the input of the LSTM layer is an influencing parameter x at the time ttAnd the predicted value h of the previous momentt-1Obtaining the predicted value h at the time t through the prediction function Ft(ii) a The function expression is as follows:
ht＝F(xt,ht-1)
3) a self-attention mechanism module is added between the three LSTM layers respectively, and the self-attention mechanism module distributes weight to the features of the hidden layers extracted from the LSTM layers, so that the features with more discriminative performance of electric load data and environmental influence data are mined;
hidden layer output sequence of penultimate layer of LSTM network at t moment is distributed from attention weight wtlThe expression is as follows:
Figure BDA0003316420310000051
wherein L ishFor the sequence length of the LSTM hidden layer output, l denotes the sequence number of the LSTM hidden layer output sequence, stlSequence l and other sequences representing the LSTN hidden layer at time tDirect similarity;
the feature h of the sequence ltlWeight w corresponding theretotlMultiplying to form a new signature sequence ht'lAnd input into the next LSTM;
step 3.3: defining optimizer, setting model training parameters
Monitoring verification loss by using the average absolute error as a loss function, and setting a self-adaptive optimizer Adam to adaptively update the learning rate in the training process; and setting the training iteration times epoch, the initial loss function and the size of batch.
Further, the predicting the electrical load in step 4 includes: ultra-short term power load prediction, and medium-long term power load prediction.
The invention has the beneficial effects that:
1) the multi-factor power load prediction method based on deep learning overcomes the defects that the current situation only faces to a single data source, the influence of complex factors such as meteorological data and the like cannot be considered, and the difference between the prediction result and the true value is large. By introducing external environmental factors such as temperature, humidity, wind power and the like which are most influence factors of the electric load and combining historical electric load data as input characteristics, the current electric load can be predicted, and the prediction precision can be effectively improved after the environmental factors are considered.
2) The invention provides a multi-factor power load forecasting method based on deep learning, and provides a method for detecting abnormal data of a power load by combining KNN and improved DBSCAN, which can effectively mine discrete values of the data value of the power load and reduce the influence of the abnormal value on a forecasting model by reducing missing or abnormal power load data.
3) The invention provides a multi-factor power load prediction method based on deep learning, and provides an improved CNN-LSTM power load prediction model, wherein a CNN characteristic extraction module is adopted to learn local characteristics of input data; then inputting the data into an LSTM sequence learning model, and extracting sequence characteristic information of input data; meanwhile, introducing a self-attention mechanism into the LSTM for learning the characteristics of an LSTM hidden layer, and realizing key purpose-made extraction by distributing different attention weights so as to improve the final prediction precision; and finally, predicting the electrical load through the Dropout layer and the FC layer. In addition, the improved CNN-LSTM-based prediction model not only can obtain higher prediction accuracy, but also can simplify the neural network construction process, and can realize end-to-end deployment, thereby promoting the research result to be rapidly put into use.
4) The multi-factor power load forecasting method based on deep learning provided by the invention realizes the forecasting of the ultra-short-term, short-term and medium-term power loads of different types of users, visualizes the forecasting result, and assists in power analysis before and after epidemic situations, power distribution analysis, power user portrait and the like. The invention can promote the digital upgrading of the power grid, meet the individual requirements of users, and realize the association analysis of industry, the dispatching of power generation, the prediction of power utilization trend, the guidance of repeated work and production and the like. The invention can be used as an independent system and can also be used as a component to be embedded into the original system of the power department, thereby saving resources to a great extent and improving the working efficiency.
Drawings
FIG. 1 is a schematic overall flow diagram of the present invention.
FIG. 2 is a data processing flow diagram of the present invention.
FIG. 3 is a schematic structural diagram of an electrical load prediction model based on the improved CNN-LSTM.
FIG. 4 is a comparison line graph of the predicted value and the actual value of the ultra-short-term electricity load for commercial electricity in a certain area according to the present invention; (a) comparing the predicted value with the true value of the five-one commercial electric load in the region in 2019; (b) compared with five commercial electric loads in the same place in 2018/2019/2020.
FIG. 5 is a line graph showing the comparison between the predicted value and the actual value of the short-term electricity load of commercial electricity in a certain area.
FIG. 6 is a comparison line graph of predicted values and actual values of medium-and long-term electricity loads for commercial electricity in a certain region according to the present invention; (a) weekly power load prediction for the site; (b) and predicting the monthly power load of the land.
Fig. 7 is a schematic diagram of an application scheme of the present invention.
Detailed Description
The invention is described in further detail below with reference to the figures and specific embodiments. The invention uses abnormal data detection, autoregressive interpolation and sequence data normalization Based on k-adjacent (KNN) algorithm and improved DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm to preprocess and normalize data by introducing external environmental factors such as temperature, humidity, wind power and the like as influence factors of electric load; then, an improved CNN-LSTM electrical load prediction model is provided, and a CNN characteristic extraction module is adopted to learn local characteristics of input data; then inputting the data into an LSTM sequence learning model, and extracting sequence characteristic information of input data; meanwhile, introducing a self-attention mechanism into the LSTM for learning the characteristics of an LSTM hidden layer, and realizing key purpose-made extraction by distributing different attention weights so as to improve the final prediction precision; and finally, forecasting the ultra-short-term power load, forecasting the short-term power load and forecasting the medium-and-long-term power load, and assisting in power analysis before and after epidemic situations, power distribution analysis, power user portrait and the like. The invention can promote the digital upgrading of the power grid, meet the individual requirements of users, and realize the association analysis of industry, the dispatching of power generation, the prediction of power utilization trend, the guidance of repeated work and production and the like.
Fig. 1 shows a flowchart of a specific implementation of the method for predicting a multi-factor electrical load based on deep learning according to the present invention, which includes: the method comprises the steps of acquiring and storing power load data and external environment data, cleaning and preprocessing data based on a KNN and improved DBSCAN algorithm, building and training based on an improved CNN-LSTM model and predicting the power load. The specific implementation steps are as follows:
step 1: and acquiring load data of different areas, years and electricity utilization types and data of external environmental factors such as temperature, humidity, wind power and the like, and storing the data in a sqlite database. The method comprises the following specific steps:
1) load data of different areas, years and electricity utilization categories are obtained from channels such as a national power grid user electricity utilization information publishing platform, the data categories comprise user numbers (desensitized), belonged regions, urban/rural networks, user categories (such as high voltage, charging piles, grid-connected high-voltage self-contained power plants, low-voltage non-residents, low-voltage residents and the like), electricity utilization categories (household electricity, commercial electricity, urban electricity, rural electricity and the like), client categories (normal electricity utilization clients and changed electricity utilization clients), voltage levels (ultrahigh voltage, high voltage and low voltage), industry categories, contract capacity, operation capacity, first power transmission date, electricity quantity freezing date, 15-minute freezing electricity quantity and the like, and the description of electricity utilization load data fields is displayed in a table 1. And the electricity load data is selected to be stored in the sqlite database, so that the data migration and the system deployment are facilitated.
TABLE 1 Electrical load data field description
Figure BDA0003316420310000071
2) The acquisition and storage of external environmental factors such as temperature, humidity, wind power and other data are completed by writing a Python crawler program. The Python crawler process involves simulating login, page acquisition and analysis, data structure design, storing and warehousing external environment data, and the like. Data of external environmental factors such as temperature, humidity, wind power and the like are acquired from an official interface of a national weather website, the data types include the current date, the region to which the data belong, the highest temperature, the lowest temperature, the average humidity, the average wind power, the weather type (the threshold range is set to be [1,4] to indicate the degree from cloudy to sunny), the current temperature of 15 minutes and whether the data is a holiday (the value is 0 or 1), and table 2 shows the description of external environmental data fields. The time of the environment data is synchronized with the time of the power load data, and similarly, the environment data is selected to be stored in the sqlite database, so that data migration and system deployment are facilitated.
Table 2 external environment data field description
Name of field Meaning of a field
WEATHER_DATE Current date
WEATHER_ADDR Region of interest
MAX_TEMP Maximum temperature
LOW_TEMP Minimum temperature
AVERAGE_HUMIDITY Average humidity
AVERAGE_WIND Mean wind force
WEATHER_TYPE Weather type (threshold [1,4]]Indicating the degree of cloudy to sunny)
R1_WEATHER Current temperature of 15 minutes
IS_HOLIDAY Whether it is a holiday (value 0 or 1)
Step 2: washing and preprocessing the power load data and the environmental impact data, comprising: abnormal data detection, autoregressive interpolation and sequence data normalization based on a k-neighbor (KNN) algorithm and an improved DBSCAN algorithm, and a data processing flow chart of the invention is schematically shown in a reference figure 2.
The method comprises the following specific steps:
1) abnormal electricity load data detection based on a k-neighbor (KNN) algorithm and an improved DBSCAN algorithm. Due to factors such as power equipment failure or human factors, the obtained user power load data has a small amount of missing or abnormal data, and the missing or abnormal data has a large influence on the prediction accuracy of the user power load, so that the missing or abnormal data needs to be detected, and normal data values need to be restored by methods such as elimination or interpolation.
The invention provides an outlier detection method based on an improved KNN algorithm. The basic rule of the KNN algorithm is to find k neighboring samples or the most similar samples of sample c in the feature space (1 < k < N, N representing the total number of samples), and if most of the k neighboring samples belong to a class, then c also belongs to this class. By calculating the average distance between the sample c and the adjacent sample as the abnormality score of the sample c, a higher abnormality score indicates that the sample c is more abnormal. In order to better reflect the abnormal condition of the power load data, after the abnormal score of the power load data at each moment is obtained by using a modified KNN algorithm, the total distance from each cluster to each k adjacent clusters is used as the final abnormal score. The electricity load data c of a certain time iiK adjacent sets N ofk(ci) Expressed as:
Figure BDA0003316420310000091
Figure BDA0003316420310000092
in the formula (I), the compound is shown in the specification,
Figure BDA0003316420310000093
denotes ciOne of the k neighboring points of dk(ci) Is ciThe average distance of k neighbors of (a),
Figure BDA0003316420310000094
denotes ciAnd its neighboring point
Figure BDA0003316420310000095
The distance of (d); electrical load data ciIs expressed as:
Figure BDA0003316420310000096
in the formula, Nk(ci) Denotes ciK sets of neighboring points;
and finally, outputting the first m clusters of the abnormal score ranking list as abnormal values of the electric load data.
Since KNN algorithm is calculating cluster ciIf k are adjacent, the abnormal cluster and the normal cluster interfere with each other, and the abnormal detection result is easy to cause false detection and missing detection. In order to solve this problem, the present invention proposes an improved cluster-based anomaly detection algorithm DBSCAN. The DBSCAN algorithm is a typical density-based clustering algorithm that ultimately yields a clustering classification result by classifying closely-connected samples into various classes, where points that do not belong to any cluster are considered to belong to outlier points. Compared with the traditional DBSCAN algorithm which adopts globally unified parameters Eps and MinPst to realize clustering, the invention provides a data partitioning improvement method based on a sliding window. The steps of the improved DBSCAN algorithm comprise three steps of parameter updating, clustering and abnormity detection.
In the process of updating the parameters, firstly, the size M of a cluster sliding window needs to be set, the average distance difference of the electric load data in the window is calculated, the former k adjacent electric loads are set as MinPst, and the Euclidean distance between the electric load dataSet to Eps. In order to alleviate the inconsistency between the electrical load data, a weight is set for each electrical load data to reduce the influence on the final clustering result. Weight w (c)i,cj) The calculation formula of (a) is as follows:
Figure BDA0003316420310000097
Figure BDA0003316420310000098
wherein, Cov (c)i,cj) Electrical load data c at time iiAnd the electric load data c at the time of jjCovariance of (a), Var (c)i) Electrical load data c at time iiVariance of (c), Var (c)j) The power load data c at time jjThe variance of (a); r (c)i,cj) Is denoted by ciAnd cjThe smaller the value, the greater the correlation; and obtaining an abnormal detection result of the electric load data by analyzing the clustering result. In the clustering process, the first-time marked abnormal point is set as a candidate abnormal point, an abnormal score is set to be added with 1 (the initial value is 0), the next candidate abnormal point is entered in the cyclic iterative clustering process, and the abnormal score is updated. If the anomaly score S is equal to the cluster number C, the anomaly point is marked.
2) For the processing of the power load missing value, an interpolation method is usually adopted to complete the missing value, and the Lagrange interpolation method is adopted to carry out interpolation on the missing power load data value. The idea of interpolation is to establish a suitable interpolation function f (x) from known points, so that the point x is not knowniThe function value f (xi) is obtained from the interpolation function f (x), so that (x) can be usedi,f(xi) Approximate instead of missing points.
The idea of Lagrange interpolation is to make n-1 polynomial y ═ a0+a1x+a2x2+L+an-1xn-1Coordinate (x) passing through n points1,y1),(x2,y2),(x3,y3),L,(xn,yn) Then the functional expression for lagrange interpolation can be expressed as:
Figure BDA0003316420310000101
wherein x isiThe i-th time, y, representing the electricity consumeriA power load indicating an i-th time of the power consumer; n represents the total number of times of the electrical load.
3) And carrying out normalization processing on the user electricity utilization data and the environment data. Since data of different types or the same type are very different, for example, the numerical values of the electrical load data and the temperature are very different, and the numerical values of the electrical load data of different types of users are also very different, if normalization processing is not performed, great influence is generated on the training of the model, and the influence of a certain parameter is hidden or amplified. Therefore, the normalization processing formula for each index data is:
Figure BDA0003316420310000102
wherein X ═ { X ═ X1,X2,X3L,XNThe normalized data represents the same type of data, such as electric load data, temperature data, or wind power, so that the normalized data has a value of Y ═ Y }1,Y2,Y3L,YNN is the total number of electrical load data or environmental impact data of a certain category.
And step 3: and (3) constructing an electrical load data prediction network model based on the improved CNN-LSMT, and FIG. 3 shows an electrical load prediction model schematic diagram based on the improved CNN-LSTM. CNN and LSTM are the most widely used deep learning techniques. The CNN model can be used to extract valuable features and can filter noise from the input data, and the LSTM network can capture sequence pattern information. The LSTM network, while capable of processing time-dependent sequence information, only utilizes the attributes provided in the training set, in contrast, CNN can be used to extract local features and the same features that occur in different regions, but it does not have features to process timing information. Therefore, the hybrid model utilizing the advantages of the two deep learning techniques can improve the prediction accuracy. In addition, the invention introduces a self-attention (self-attention) mechanism to improve the CNN-LSTM, and can perform feature reconstruction on the hidden layer of the LSTM.
Therefore, the specific steps are as follows:
and 3.1) preparing training data of an electrical load prediction model, wherein the electrical load data and the environmental data obtained by processing in the step 2) are used as training data, for example, short-term (daily) electrical load prediction is taken as an example, the training data are preprocessed data of electrical load and environmental factors such as the highest temperature, the lowest temperature, the average humidity and the like of a certain user in 365 days in one year, and the electrical load prediction model is a neural network. Firstly loading a data set, dividing a training set, a verification set and a test set according to the ratio of 7:2:1, then respectively constructing a Dataloader for data reading on the training set and the test set, calculating the data of each batch when a model is trained, loading the data into a memory by the size of the batch, disordering the data of each batch, and improving the robustness of model training, wherein the size of the batch is set to be 10.
And 3.2) building an electrical load prediction model, wherein the deep learning model in the invention adopts a Python language experiment, and the improved electrical load prediction model of CNN-LSTM is realized based on a Pythrch deep learning library. The improved CNN-LSTM power load prediction model comprises a CNN feature learning module, an LSTM sequence learning module and a self-attention (self-attention) mechanism module.
The CNN feature learning module consists of three one-dimensional convolution layers, wherein a Max painting layer and a ReLu layer are added between two continuous convolution layers. The feature of the normalized power load data and the feature of the environment data are learned by introducing convolution operation, and the feature graph output by the convolution layer has a limit, namely the feature graph can track the accurate position of the features of the input data, and can obtain more discriminative features in the input data. A MaxPooling layer is usually added after the convolutional layer to alleviate the limitation of invariance of the generated feature map, and the activation function ReLu is used to enhance the ability of the model to learn complex structures, thereby reducing the overall computational load and making the network easier to train.
When the power load is predicted, the time sequence correlation of load data is considered fully, compared with the traditional recurrent neural network, the LSTM can accurately learn the long-term dependence relationship in the time sequence and is suitable for learning the power load data with a long period, so that the method can obtain higher prediction precision by using the LSTM to predict the power load. Input as the influencing parameter x at time ttAnd the predicted value h of the previous momentt-1Obtaining the predicted value h at the time t through the prediction function Ft. The functional expression is as follows:
ht＝F(xt,ht-1)
the LSTM enhances the ability of the LSTM to handle long sequence dependency problems by adding memory and control gates to the recurrent neural network, and shows better performance in predicting the electrical load.
In the LSTM sequence learning module, the present invention uses three LSTM layers, each layer containing twenty neurons. The first two LSTM layers output the complete sequence of hidden states, while at the last LSTM layer, the last time step of the hidden state is output.
In order to fully mine the final prediction capability of the features of the hidden layer in the LSTM on the model, a self-attention module is respectively added among three LSTMs of the sequence learning module, and a self-attention mechanism distributes weight to the features of the hidden layer extracted by the LSTM, so that the features with higher discriminability of the electric load data and the environmental data are mined, and the final model prediction accuracy is improved. Hidden layer output sequence of penultimate layer of LSTM network at t moment is distributed from attention weight wtlThe expression is as follows:
Figure BDA0003316420310000121
wherein L ishSequence length output for LSTM hidden layer, l denotes LSTM hiddenSequence number of hidden layer output sequence, stlRepresenting the direct similarity of the sequence l of the LSTN hidden layer at the time t and other sequences;
the feature h of the sequence ltlWeight w corresponding theretotlMultiplying to form a new signature sequence ht'lAnd input into the next LSTM. A self-attribute mechanism is introduced to give weight to the characteristics of the hidden layer of the LSTM, so that key information in the sequence can be effectively acquired, and the accuracy and efficiency of prediction are improved.
In the development of any deep learning model, the dropout layer includes randomly selecting neurons and deactivating some of them during the training process to prevent overfitting of the model. In the invention, a dropout layer is added between the CNN feature extraction block and the LSTM sequence learning to prevent overfitting. The output of the LSTM sequence learning block is also connected to a dropout layer, followed by a fully connected layer (FC) to produce the final output.
And 3.3) defining an optimizer and setting model training parameters. The ratio of training data, validation data and test data in the model training input data was 7:2:1, and the validation loss was monitored using the Mean Absolute Error (MAE) as a loss function. The optimizer of the model is set as a self-adaptive optimizer Adam, the initial learning rate is set as 0.001, and Adam can update the learning rate in a self-adaptive manner in the training process, so that the convergence speed of the model is high, and the parameters are adjusted more easily. In addition, the training iteration number epoch of the model is set to 700, the initial loss function of each iteration is set to 0, and the size of the batch is set to 10.
And 4, step 4: the prediction of the power load mainly comprises the following steps: the ultra-short term electricity load prediction, the short term electricity load prediction and the medium-long term electricity load prediction refer to fig. 4, fig. 5 and fig. 6 respectively show comparative line graphs of the ultra-short term, the short term and the medium-long term electricity load prediction and the user real electricity load.
Fig. 4 (a) is a comparison between a predicted value and a true value of a five-one commercial electric load in a certain place in 2019; by analyzing the real value and the predicted value of the electricity utilization in a certain day, the predicted value of the commercial electricity utilization in the certain day can be obtained and basically reflects the real electricity utilization situation, and the scheduling guidance can be provided for the power plant. (b) The comparison of the five-one commercial electric loads of the place in 2018/2019/2020 shows that the five-one commercial electric loads of the place in 2018/2019/2020 are analyzed to obtain that the influence of epidemic situations on the five-one commercial electric loads of the place in 2020 is larger than that on the commercial electric loads of the place in the first two years.
Fig. 5 is a comparison line drawing of a short-term electricity load predicted value and a real value of commercial electricity in a certain region, a temperature factor is introduced to predict the commercial electricity in the region in 2019, the predicted value can basically reflect the real load electricity utilization trend and the change trend of the load temperature, but an abnormal condition occurs around 2 months and 5 months, because most of the commercial electricity in the region is in a vacation state during the spring festival.
Fig. 6 is a comparison line graph of the predicted value and the actual value of the medium-long term electricity load of the commercial electricity in a certain region. (a) Weekly power load prediction for the site; through comparing the electricity load value of the commercial electricity consumption every week with the predicted value, the predicted value can basically reflect the change trend of the real value. (b) For the monthly power load prediction of the place, the predicted value can basically reflect the change trend of the real value by comparing the real value with the predicted value of the monthly power load of the commercial power.
The method comprises the following specific steps:
taking commercial power utilization of a certain city as an example, the design and implementation of the power load prediction module mainly predict the power load from multiple dimensions of ultra-short term, short term and medium term from the power generation department of the power plant, so that the power generation department of the power plant can conveniently master the commercial power utilization conditions of the certain city every day, every week and every month, and can make effective power load scheduling. For example, acquiring power load data of multiple dimensions of power utilization time, day, week and month of a certain market in 2019 and environmental data such as maximum temperature, minimum temperature, average humidity, average wind power size, weather type and the like of each day, preprocessing and standardizing the data to be used as input of an LSTM model, outputting predicted power load trends of multiple dimensions of power utilization time, day, week and month of the certain market in 2019, reflecting the accuracy of the model on power load prediction by comparing a real value with a predicted value, and introducing the change trend of factors such as temperature and the comparison of the power load trends to reflect the influence of the environmental factors on the power load size.
The method is mainly applied to the power department, such as guiding a power plant to be used for power generation and power dispatching, assisting the power management department to analyze power utilization conditions, mining user power utilization information, being applicable to research and development departments, mining user power utilization rules and improving the work efficiency of the power department, and an application scheme of the method is shown by referring to fig. 7. The power load of a power grid is influenced by factors such as regions, time, temperature, power type, voltage type and the like, and the multi-factor power load prediction system based on deep learning is designed by preprocessing and standardizing data by using abnormal data detection, autoregressive interpolation and sequence data normalization based on a k-neighbor (KNN) algorithm and an improved DBSCAN algorithm; then, an improved CNN-LSTM electrical load prediction model is provided, and a CNN characteristic extraction module is adopted to learn local characteristics of input data; then inputting the data into an LSTM sequence learning model, and extracting sequence characteristic information of input data; meanwhile, introducing a self-attention mechanism into the LSTM for learning the characteristics of an LSTM hidden layer, and realizing key purpose-made extraction by distributing different attention weights so as to improve the final prediction precision; and finally, forecasting the ultra-short-term power load, forecasting the short-term power load and forecasting the medium-and-long-term power load, and assisting in power analysis before and after epidemic situations, power distribution analysis, power user portrait and the like. The method can be applied to the power department to predict the power utilization trend of users, reasonably plan the periodic power utilization, help the power department to schedule the power generation amount of the power plant, and help to guide the re-work and re-production of the industry and the association analysis of the industry.

Claims (7)

1. A multi-factor power load prediction method based on deep learning is characterized by comprising the following steps:
step 1: acquiring power load data including different areas, years and power utilization categories and environmental influence data including temperature, humidity and wind power, and storing the data in a sqlite database;
step 2: the method for cleaning and preprocessing the power load data and the environmental impact data comprises the following steps: abnormal data detection, autoregressive interpolation and sequence data normalization based on a k-proximity algorithm and an improved DBSCAN algorithm;
and step 3: introducing a self-attention mechanism, performing feature reconstruction on a hidden layer of the LSTM to realize end-to-end deployment, constructing an electrical load data prediction neural network model based on the improved CNN-LSMT, and using the electrical load data and the environmental impact data obtained by processing in the step (2) as a training set and a test set;
and 4, step 4: and carrying out the electric load prediction through the electric load data prediction neural network model of the improved CNN-LSMT.
2. The deep learning-based multi-factor power load prediction method according to claim 1, wherein the power load data includes a user number, a region to which the user belongs, an urban/rural network, a user classification, a power category, a customer category, a voltage class, an industry category, a contract capacity, an operating capacity, a first power transmission date, a frozen power date, and a frozen power of 15 minutes; the environmental impact data comprises the current date, the region, the highest temperature, the lowest temperature, the average humidity, the average wind power, the weather type, the current temperature of 15 minutes and whether the current temperature is a holiday or not; and the time of the environmental impact data is synchronized with the time of the electrical load data.
3. The method for predicting the electrical load with the multiple factors based on the deep learning according to claim 1, wherein in the step 2, the abnormal data detection based on the k-neighbor algorithm and the improved DBSCAN algorithm is specifically as follows:
step 2.1: defining the average distance between a sample and an adjacent sample as the abnormal score of the sample, obtaining the abnormal score of the electricity load data at each moment by using a modified KNN algorithm, and taking the total distance from each cluster to each k adjacent samples as the final abnormal score;
load data c of electricity consumption at a certain time iiK adjacent sets N ofk(ci) Expressed as:
Figure FDA0003316420300000011
Figure FDA0003316420300000012
in the formula (I), the compound is shown in the specification,
Figure FDA0003316420300000013
denotes ciOne of the k neighboring points of dk(ci) Is ciThe average distance of k neighbors of (a),
Figure FDA0003316420300000014
denotes ciAnd its neighboring point
Figure FDA0003316420300000015
The distance of (d);
electrical load data ciIs expressed as:
Figure FDA0003316420300000016
in the formula, Nk(ci) Denotes ciK sets of neighboring points;
finally, outputting the first m clusters of the abnormal score ranking list as abnormal values of the electric load data;
step 2.2: adopting an improved cluster-based anomaly detection algorithm DBSCAN, firstly utilizing local parameters to realize density clustering of data of small samples, then carrying out iterative clustering on local clustering results to realize a final global clustering result, and marking points which do not belong to any cluster as anomalous points; the method specifically comprises the following steps:
step a) updating parameters:
setting the size M of a cluster sliding window, calculating the average distance difference of electric load data in the window, setting the former k adjacent electric loads as MinPst, and setting the Euclidean distance between the electric load data as Eps;
setting a weight for each load data to reduce the impact on the final clustering result, weight w (c)i,cj) The calculation formula of (a) is as follows:
Figure FDA0003316420300000021
Figure FDA0003316420300000022
wherein, Cov (c)i,cj) Electrical load data c at time iiAnd the electric load data c at the time of jjCovariance of (a), Var (c)i) Electrical load data c at time iiVariance of (c), Var (c)j) The power load data c at time jjThe variance of (a); r (c)i,cj) Is denoted by ciAnd cjThe smaller the value, the greater the correlation;
step b) obtaining an abnormal detection result of the electric load data by analyzing the clustering result:
in the clustering process, setting the abnormal point marked for the first time as a candidate abnormal point, setting the abnormal score plus 1, entering the next candidate abnormal point in the cyclic iterative clustering process, and updating the abnormal score; if the anomaly score S is equal to the cluster number C, the anomaly point is marked.
4. The method for predicting the electrical load according to claim 1, wherein in the step 2, the autoregressive interpolation is a method comprising:
completing the missing electric load data value by adopting a Lagrange interpolation method so that the polynomial y of n-1 is a0+a1x+a2x2+…+an-1xn-1Coordinate (x) passing through n points1,y1),(x2,y2),(x3,y3),…,(xn,yn) Then, thenThe functional expression of the lagrange interpolation is expressed as:
Figure FDA0003316420300000023
in the formula, xiAnd xjI and j time instants, y, representing the power consumers, respectivelyiA power load indicating an i-th time of the power consumer; n represents the total number of times of the electrical load.
5. The method for predicting the electrical load according to claim 1, wherein in the step 2, the processing formula of the normalization of the sequence data is as follows:
Figure FDA0003316420300000031
wherein X is { X ═ X1,X2,X3…,XNRepresents the same category of electrical load data or environmental impact data; the normalized value is Y ═ Y1,Y2,Y3…,YN}，l∈[1,N]And N is the total number of certain types of electrical load data or environmental impact data.
6. The method for predicting the multi-factor electrical load based on the deep learning of claim 1, wherein the step 3 specifically comprises:
step 3.1: for the loading and preparation of network model data:
loading a data set, dividing a training set, a verification set and a test set, constructing a Dataloader as a data reader for the training set and the test set respectively, calculating data of each batch during model training, loading the data into a memory by the Dataloader according to the size of the batch, and disordering the data of each batch to improve the robustness of the model training;
step 3.2: constructing an electrical load data prediction neural network model based on the improved CNN-LSMT:
the neural network model comprises a CNN feature learning module, an LSTM sequence learning module and an attention mechanism module:
1) the CNN feature learning module comprises three one-dimensional convolution layers, wherein a Max painting layer and a ReLu layer are added between two continuous convolution layers; learning the characteristics of the normalized power load data and the normalized environmental impact data through convolution operation, and using the characteristics as a characteristic diagram output by the convolution layer; adding a MaxPholing layer to relieve the limitation of invariance of the generated feature map, and activating a function ReLu to enhance the capability of the model to learn a complex structure;
2) the LSTM sequence learning module comprises three LSTM layers, each layer comprising twenty neurons; the first two LSTM layers output the complete sequence of the hidden state, and the last LSTM layer outputs the last time step of the hidden state;
the input of the LSTM layer is an influencing parameter x at the time ttAnd the predicted value h of the previous momentt-1Obtaining the predicted value h at the time t through the prediction function Ft(ii) a The function expression is as follows:
ht＝F(xt,ht-1)
3) a self-attention mechanism module is added between the three LSTM layers respectively, and the self-attention mechanism module distributes weight to the features of the hidden layers extracted from the LSTM layers, so that the features with more discriminative performance of electric load data and environmental influence data are mined; hidden layer output sequence of penultimate layer of LSTM network at t moment is distributed from attention weight wtlThe expression is as follows:
Figure FDA0003316420300000032
in the formula, LhFor the sequence length of the LSTM hidden layer output, l denotes the sequence number of the LSTM hidden layer output sequence, stlRepresenting the direct similarity of the sequence l of the LSTN hidden layer at the time t and other sequences;
the feature h of the sequence ltlWeight w corresponding theretotlMultiplying to form a new signature sequence ht'lAnd input into the next LSTM;
step 3.3: defining an optimizer, setting model training parameters:
monitoring verification loss by using the average absolute error as a loss function, and setting a self-adaptive optimizer Adam to adaptively update the learning rate in the training process; and setting the training iteration times epoch, the initial loss function and the size of batch.
7. The method for predicting the electrical load based on the multi-factor deep learning of claim 1, wherein the predicting the electrical load in the step 4 comprises: ultra-short term power load prediction, and medium-long term power load prediction.
CN202111232185.5A 2021-10-22 2021-10-22 Multi-factor power load prediction method based on deep learning Active CN113962364B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111232185.5A CN113962364B (en) 2021-10-22 2021-10-22 Multi-factor power load prediction method based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111232185.5A CN113962364B (en) 2021-10-22 2021-10-22 Multi-factor power load prediction method based on deep learning

Publications (2)

Publication Number Publication Date
CN113962364A true CN113962364A (en) 2022-01-21
CN113962364B CN113962364B (en) 2023-04-18

Family

ID=79466128

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111232185.5A Active CN113962364B (en) 2021-10-22 2021-10-22 Multi-factor power load prediction method based on deep learning

Country Status (1)

Country Link
CN (1) CN113962364B (en)

Cited By (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114638440A (en) * 2022-05-17 2022-06-17 国网湖北省电力有限公司经济技术研究院 Charging load ultra-short-term prediction method based on charging pile utilization degree
CN114676941A (en) * 2022-05-30 2022-06-28 国网天津市电力公司经济技术研究院 Electric-thermal load combined self-adaptive prediction method and device for park comprehensive energy system
CN114970693A (en) * 2022-05-11 2022-08-30 国网上海市电力公司 Charging pile user portrait method based on federal learning
CN115018119A (en) * 2022-04-26 2022-09-06 河北大学 Power load prediction method and system
CN115271463A (en) * 2022-07-29 2022-11-01 昂顿科技（上海）有限公司 Intelligent energy management system
CN115412923A (en) * 2022-10-28 2022-11-29 河北省科学院应用数学研究所 Multi-source sensor data credible fusion method, system, equipment and storage medium
CN115481788A (en) * 2022-08-31 2022-12-16 北京建筑大学 Load prediction method and system for phase change energy storage system
CN115587617A (en) * 2022-10-17 2023-01-10 国网黑龙江省电力有限公司 Method and device for capturing sequence information of long-time sequence and method and device for predicting short-term power consumption
CN115936185A (en) * 2022-11-15 2023-04-07 国网江苏省电力有限公司苏州供电分公司 Short-term power load and carbon emission prediction method and system based on DCNN-LSTM-AE-AM
CN116128124A (en) * 2023-01-09 2023-05-16 北京建筑大学 Building energy consumption prediction method based on abnormal energy value processing and time sequence decomposition
CN116205666A (en) * 2022-12-22 2023-06-02 国网湖北省电力有限公司宜昌供电公司 RACNet-based multivariable power load prediction method
CN116402483A (en) * 2023-06-09 2023-07-07 国网山东省电力公司兰陵县供电公司 Online monitoring method and system for carbon emission of park
CN116523277A (en) * 2023-07-05 2023-08-01 北京观天执行科技股份有限公司 Intelligent energy management method and system based on demand response
CN116562168A (en) * 2023-06-09 2023-08-08 岳阳融盛实业有限公司 Electric power informatization data mining system and method based on deep learning
CN116932487A (en) * 2023-09-15 2023-10-24 北京安联通科技有限公司 Quantized data analysis method and system based on data paragraph division
CN117009861A (en) * 2023-10-08 2023-11-07 湖南国重智联工程机械研究院有限公司 Hydraulic pump motor life prediction method and system based on deep learning
CN117035837A (en) * 2023-10-09 2023-11-10 广东电力交易中心有限责任公司 Method for predicting electricity purchasing demand of power consumer and customizing retail contract
CN117111540A (en) * 2023-10-25 2023-11-24 南京德克威尔自动化有限公司 Environment monitoring and early warning method and system for IO remote control bus module
CN117350439A (en) * 2023-11-16 2024-01-05 华南理工大学 Energy aggregation service provider load prediction method and system based on transverse federal learning
CN117689278A (en) * 2024-02-04 2024-03-12 新疆盛诚工程建设有限责任公司 Construction quality intelligent management system and method
CN117878927A (en) * 2024-03-11 2024-04-12 国网黑龙江省电力有限公司绥化供电公司 Power system load trend analysis method based on time sequence analysis
CN117895659A (en) * 2024-03-14 2024-04-16 山东理工大学 Automatic scheduling method and system for smart power grid
CN117895659B (en) * 2024-03-14 2024-05-31 山东理工大学 Automatic scheduling method and system for smart power grid

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109934392A (en) * 2019-02-28 2019-06-25 武汉大学 A kind of micro-capacitance sensor short-term load forecasting method based on deep learning
CN110659779A (en) * 2019-09-26 2020-01-07 国网湖南省电力有限公司 Power distribution system network loss prediction method based on long-term and short-term memory network
US20200143460A1 (en) * 2018-11-02 2020-05-07 University-Industry Cooperation Group Of Kyung-Hee University System of management of energy trading and method of the same

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200143460A1 (en) * 2018-11-02 2020-05-07 University-Industry Cooperation Group Of Kyung-Hee University System of management of energy trading and method of the same
CN109934392A (en) * 2019-02-28 2019-06-25 武汉大学 A kind of micro-capacitance sensor short-term load forecasting method based on deep learning
CN110659779A (en) * 2019-09-26 2020-01-07 国网湖南省电力有限公司 Power distribution system network loss prediction method based on long-term and short-term memory network

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
VASILEIOS TOSOUNIDIS: "Wind Power Forecasting with Deep Learning Methods" *
XINYI WANG 等: "Self-Attention based Neural Network for Predicting RNA-Protein Binding Sites" *
闫建荣: "基于数据增强的RNA-蛋白质相互作用预测研究" *

Cited By (35)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115018119A (en) * 2022-04-26 2022-09-06 河北大学 Power load prediction method and system
CN114970693B (en) * 2022-05-11 2024-05-07 国网上海市电力公司 Charging pile user portrait method based on federal learning
CN114970693A (en) * 2022-05-11 2022-08-30 国网上海市电力公司 Charging pile user portrait method based on federal learning
CN114638440B (en) * 2022-05-17 2022-08-09 国网湖北省电力有限公司经济技术研究院 Charging load ultra-short-term prediction method based on charging pile utilization degree
CN114638440A (en) * 2022-05-17 2022-06-17 国网湖北省电力有限公司经济技术研究院 Charging load ultra-short-term prediction method based on charging pile utilization degree
CN114676941A (en) * 2022-05-30 2022-06-28 国网天津市电力公司经济技术研究院 Electric-thermal load combined self-adaptive prediction method and device for park comprehensive energy system
CN114676941B (en) * 2022-05-30 2022-09-30 国网天津市电力公司经济技术研究院 Electric-thermal load combined self-adaptive prediction method and device for park comprehensive energy system
CN115271463A (en) * 2022-07-29 2022-11-01 昂顿科技（上海）有限公司 Intelligent energy management system
CN115481788A (en) * 2022-08-31 2022-12-16 北京建筑大学 Load prediction method and system for phase change energy storage system
CN115481788B (en) * 2022-08-31 2023-08-25 北京建筑大学 Phase change energy storage system load prediction method and system
CN115587617A (en) * 2022-10-17 2023-01-10 国网黑龙江省电力有限公司 Method and device for capturing sequence information of long-time sequence and method and device for predicting short-term power consumption
CN115412923A (en) * 2022-10-28 2022-11-29 河北省科学院应用数学研究所 Multi-source sensor data credible fusion method, system, equipment and storage medium
CN115936185A (en) * 2022-11-15 2023-04-07 国网江苏省电力有限公司苏州供电分公司 Short-term power load and carbon emission prediction method and system based on DCNN-LSTM-AE-AM
CN116205666A (en) * 2022-12-22 2023-06-02 国网湖北省电力有限公司宜昌供电公司 RACNet-based multivariable power load prediction method
CN116128124A (en) * 2023-01-09 2023-05-16 北京建筑大学 Building energy consumption prediction method based on abnormal energy value processing and time sequence decomposition
CN116402483B (en) * 2023-06-09 2023-08-18 国网山东省电力公司兰陵县供电公司 Online monitoring method and system for carbon emission of park
CN116562168B (en) * 2023-06-09 2024-05-14 岳阳融盛实业有限公司 Electric power informatization data mining system and method based on deep learning
CN116402483A (en) * 2023-06-09 2023-07-07 国网山东省电力公司兰陵县供电公司 Online monitoring method and system for carbon emission of park
CN116562168A (en) * 2023-06-09 2023-08-08 岳阳融盛实业有限公司 Electric power informatization data mining system and method based on deep learning
CN116523277A (en) * 2023-07-05 2023-08-01 北京观天执行科技股份有限公司 Intelligent energy management method and system based on demand response
CN116523277B (en) * 2023-07-05 2023-10-20 北京观天执行科技股份有限公司 Intelligent energy management method and system based on demand response
CN116932487B (en) * 2023-09-15 2023-11-28 北京安联通科技有限公司 Quantized data analysis method and system based on data paragraph division
CN116932487A (en) * 2023-09-15 2023-10-24 北京安联通科技有限公司 Quantized data analysis method and system based on data paragraph division
CN117009861B (en) * 2023-10-08 2023-12-15 湖南国重智联工程机械研究院有限公司 Hydraulic pump motor life prediction method and system based on deep learning
CN117009861A (en) * 2023-10-08 2023-11-07 湖南国重智联工程机械研究院有限公司 Hydraulic pump motor life prediction method and system based on deep learning
CN117035837B (en) * 2023-10-09 2024-01-19 广东电力交易中心有限责任公司 Method for predicting electricity purchasing demand of power consumer and customizing retail contract
CN117035837A (en) * 2023-10-09 2023-11-10 广东电力交易中心有限责任公司 Method for predicting electricity purchasing demand of power consumer and customizing retail contract
CN117111540A (en) * 2023-10-25 2023-11-24 南京德克威尔自动化有限公司 Environment monitoring and early warning method and system for IO remote control bus module
CN117111540B (en) * 2023-10-25 2023-12-29 南京德克威尔自动化有限公司 Environment monitoring and early warning method and system for IO remote control bus module
CN117350439A (en) * 2023-11-16 2024-01-05 华南理工大学 Energy aggregation service provider load prediction method and system based on transverse federal learning
CN117689278A (en) * 2024-02-04 2024-03-12 新疆盛诚工程建设有限责任公司 Construction quality intelligent management system and method
CN117878927A (en) * 2024-03-11 2024-04-12 国网黑龙江省电力有限公司绥化供电公司 Power system load trend analysis method based on time sequence analysis
CN117878927B (en) * 2024-03-11 2024-05-28 国网黑龙江省电力有限公司绥化供电公司 Power system load trend analysis method based on time sequence analysis
CN117895659A (en) * 2024-03-14 2024-04-16 山东理工大学 Automatic scheduling method and system for smart power grid
CN117895659B (en) * 2024-03-14 2024-05-31 山东理工大学 Automatic scheduling method and system for smart power grid

Also Published As

Publication number Publication date
CN113962364B (en) 2023-04-18

Similar Documents

Publication Publication Date Title
CN113962364B (en) Multi-factor power load prediction method based on deep learning
Azeem et al. Electrical load forecasting models for different generation modalities: a review
CN106951611A (en) A kind of severe cold area energy-saving design in construction optimization method based on user&#39;s behavior
US20210326696A1 (en) Method and apparatus for forecasting power demand
CN112465256A (en) Building power consumption prediction method and system based on Stacking model fusion
Liu et al. Edge sensing data-imaging conversion scheme of load forecasting in smart grid
CN115130741A (en) Multi-model fusion based multi-factor power demand medium and short term prediction method
Li et al. Decomposition integration and error correction method for photovoltaic power forecasting
CN113554466A (en) Short-term power consumption prediction model construction method, prediction method and device
Zhang et al. A novel sequence to sequence data modelling based CNN-LSTM algorithm for three years ahead monthly peak load forecasting
CN115860177A (en) Photovoltaic power generation power prediction method based on combined machine learning model and application thereof
CN113591368A (en) Comprehensive energy system multi-energy load prediction method and system
Chen et al. MultiCycleNet: multiple cycles self-boosted neural network for short-term electric household load forecasting
CN115238854A (en) Short-term load prediction method based on TCN-LSTM-AM
Al-Ja’afreh et al. An enhanced CNN-LSTM based multi-stage framework for PV and load short-term forecasting: DSO scenarios
Eren et al. A comprehensive review on deep learning approaches for short-term load forecasting
CN114611757A (en) Electric power system short-term load prediction method based on genetic algorithm and improved depth residual error network
Mishra et al. Performance evaluation of prophet and STL-ETS methods for load forecasting
CN114595861A (en) MSTL (modeling, transformation, simulation and maintenance) and LSTM (least Square TM) model-based medium-and-long-term power load prediction method
Yang et al. Decomposition strategy and attention-based long short-term memory network for multi-step ultra-short-term agricultural power load forecasting
CN117390550A (en) Low-carbon park carbon emission dynamic prediction method and system considering emission training set
Zuo Integrated forecasting models based on LSTM and TCN for short-term electricity load forecasting
CN116485582A (en) Heat supply optimization regulation and control method and device based on deep learning
CN116090635A (en) Meteorological-driven new energy generation power prediction method
CN113762591B (en) Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning

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