CN116432861A - Comprehensive energy system source charge price multitasking combined prediction method and system - Google Patents

Comprehensive energy system source charge price multitasking combined prediction method and system Download PDF

Info

Publication number
CN116432861A
CN116432861A CN202310519941.5A CN202310519941A CN116432861A CN 116432861 A CN116432861 A CN 116432861A CN 202310519941 A CN202310519941 A CN 202310519941A CN 116432861 A CN116432861 A CN 116432861A
Authority
CN
China
Prior art keywords
prediction
price
sharing
feature
energy system
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
CN202310519941.5A
Other languages
Chinese (zh)
Inventor
李珂
杨帆
牟宇宸
张承慧
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.)
Shandong University
Original Assignee
Shandong 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 Shandong University filed Critical Shandong University
Priority to CN202310519941.5A priority Critical patent/CN116432861A/en
Publication of CN116432861A publication Critical patent/CN116432861A/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
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • 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
    • G06N3/048—Activation functions
    • 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
    • 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
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
    • G06Q10/06315—Needs-based resource requirements planning or analysis
    • 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
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • 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
    • 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)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Economics (AREA)
  • General Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Business, Economics & Management (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Marketing (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Tourism & Hospitality (AREA)
  • Artificial Intelligence (AREA)
  • Development Economics (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Game Theory and Decision Science (AREA)
  • Educational Administration (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Primary Health Care (AREA)
  • Water Supply & Treatment (AREA)
  • Public Health (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a comprehensive energy system source charge and price multitask combined prediction method and a system, which analyze the space-time coupling characteristics among all uncertainty factors affecting a comprehensive energy system from two angles of cross correlation and autocorrelation; extracting features from the two correlation analysis results; extracting the extracted features for the second time by using a channel attention and time sequence attention mechanism; the invention can analyze the correlation of two uncertainties of an energy level and a price level and develop the joint prediction under the background of multiple uncertainties of a comprehensive energy system.

Description

Comprehensive energy system source charge price multitasking combined prediction method and system
Technical Field
The invention belongs to the technical field of comprehensive energy system planning, and relates to a comprehensive energy system source charge and price multitask combined prediction method and a system.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
The permeability of new energy sources such as wind power, photovoltaic and the like at the energy supply side of the comprehensive energy system is gradually improved, flexible loads such as energy-side electric vehicles, energy storage batteries and the like are used for being largely connected, more private capital is encouraged to enter the energy market to participate in competitive transaction, uncertainty factors in the system are obviously increased, and a plurality of links such as energy generation, transaction and consumption are covered. Accurate uncertainty prediction is a prerequisite for ensuring efficient operation of integrated energy systems.
The traditional prediction method is used for independently predicting three kinds of uncertainties, namely new energy power at the energy supply side, energy consumption side load and energy price at the market side. The model is also mature, and results such as point prediction, interval prediction, probability prediction and the like are obtained by adopting a time sequence analysis or machine learning method. However, there is a complex coupling relationship between various random variables in the integrated energy system, and in this context, it is important to comprehensively consider and analyze the correlation between random variables and their influencing factors, and fully extract the spatial and temporal coupling characteristics thereof, so as to implement accurate prediction under the condition of multiple variables. Compared with a single-variable independent prediction method, the multi-element random variable combined prediction is developed, so that the internal relation of uncertain elements in the comprehensive energy system can be further excavated, and the prediction precision and the prediction efficiency are improved.
In short, the traditional independent prediction method faces the problems of low efficiency and difficult accurate prediction in the face of complex coupling characteristics among uncertainties in an energy-price layer in a comprehensive energy system. Further, research on the interaction between the user behavior habit and the load demand response on the energy utilization side and the energy price on the market level is carried out, and a proper prediction model is built, so that the realization of accurate combined prediction of the comprehensive energy system source and the load price is still challenging.
Disclosure of Invention
In order to solve the problems, the invention provides a comprehensive energy system source charge multi-task combined prediction method and a system, and the method can analyze the correlation of two uncertainties of an energy layer and a price layer and develop combined prediction under the background of multiple uncertainties of a comprehensive energy system.
According to some embodiments, the present invention employs the following technical solutions:
a comprehensive energy system source charge price multitasking joint prediction method comprises the following steps:
analyzing the space-time coupling characteristics among all uncertainty factors affecting the comprehensive energy system from the two angles of cross correlation and autocorrelation;
constructing a training data set according to the two correlation analysis results and extracting features;
extracting the extracted features for the second time by using a channel attention and time sequence attention mechanism;
and carrying out feature sharing on the data after secondary extraction by using a parameter sharing learning mechanism, wherein the multitask learning method takes price prediction as a main task, source and load prediction as auxiliary tasks, coupling among corresponding loads of various tasks adopts a hard sharing mechanism, and soft sharing mechanisms are adopted among different types of tasks, so that the auxiliary tasks share information to the main task, and a joint prediction result is obtained.
As an alternative embodiment, the correlation analysis between the multiple load, the new energy power generation, the energy price, the time and the meteorological factors is performed by using the Pearson product difference correlation coefficient, the Spearman rank correlation coefficient and the Kendall rank correlation coefficient, respectively.
Alternatively, the time sequence characteristics of each factor are analyzed by using the autocorrelation coefficients to determine the data time length adopted by the prediction.
As an alternative embodiment, the specific process of extracting features from the two correlation analysis results includes: extracting features of input data through a convolution layer, wherein the convolution layer regularly traverses the input data, performs matrix element multiplication summation on the input data and superimposes deviation values;
transmitting the feature map output by the convolution layer to a pooling layer by utilizing activation of a ReLU function, sliding a window of the pooling layer on all areas of the input according to the stride size, traversing each position through the window, and calculating the output;
and (5) iterating the process to obtain the final output characteristic.
As an alternative embodiment, the specific process of performing secondary extraction on the extracted features by using the channel attention and time sequence attention mechanism includes: the channel attention module is connected with the time sequence attention module in series, the input characteristics are used for obtaining a channel attention matrix through the channel attention module, the matrix is multiplied by the original image to obtain characteristics and used as the input of the time sequence attention module, the time attention matrix is obtained through the time sequence attention module, and then the time attention matrix is multiplied by the original image to obtain output characteristics.
As an optional implementation manner, the processing procedure of the channel attention module includes that an input feature sequence passes through two parallel MaxPool layers and an AvgPool layer, feature sequence dimensions are compressed, the channel number is compressed and then expanded to the original channel number, and two activated results are obtained through a ReLU activation function; and adding the two output results element by element, and obtaining a channel attention matrix through a sigmoid activation function.
The processing procedure of the time sequence attention module comprises the following steps: splicing the data of different channels according to the time dimension by the feature sequence processed by the channel attention module to obtain the feature sequence; and extracting features by a one-dimensional convolution layer, converting the dimension of the feature sequence into the original dimension, and obtaining a time sequence attention matrix through an activation function Sigmoid.
As an alternative implementation mode, the LSTM network is utilized for feature sharing, the thermoelectric load is directly subjected to hard sharing through the LSTM network, wind power, load features and electricity price features are subjected to soft sharing through weight sum among different LSTM network layers, and auxiliary tasks share information to a main task when outputting.
A comprehensive energy system source charge price multitasking joint prediction system comprising:
a correlation analysis module configured to analyze a spatiotemporal coupling characteristic between uncertainty factors affecting the integrated energy system from both a cross correlation and an autocorrelation perspective;
the feature extraction module is configured to perform feature extraction on two correlation analysis results;
the secondary extraction module is configured to perform secondary extraction on the extracted features by using a channel attention and time sequence attention mechanism;
the feature sharing module is configured to construct a double-layer feature sharing model by utilizing a hard sharing mechanism and a soft sharing mechanism, and perform feature sharing on the data after secondary extraction, wherein the double-layer feature sharing model adopts the hard sharing mechanism according to the division of the predicted task types, the coupling among the corresponding loads of various tasks adopts the hard sharing mechanism, and adopts the soft sharing mechanism among different types of tasks, so that auxiliary tasks share information to main tasks, and a joint prediction result is obtained.
A computer readable storage medium having stored therein a plurality of instructions adapted to be loaded by a processor of a terminal device and to perform the steps in the method.
A terminal device comprising a processor and a computer readable storage medium, the processor configured to implement instructions; the computer readable storage medium is for storing a plurality of instructions adapted to be loaded by a processor and to perform the steps in the method.
Compared with the prior art, the invention has the beneficial effects that:
the invention carries out correlation analysis and joint prediction on three kinds of uncertainty of source-charge-price under the background of multiple uncertainties of a comprehensive energy system. The proposed multi-task learning model classifies prediction targets according to different correlation degrees and attribute types, and in the double-layer feature sharing module, different sharing mechanisms are adopted according to different task types to distinguish an outer layer from an inner layer, so that reasonable sharing of important features is promoted, the generalization capability of the model is improved, and multi-uncertainty efficient and accurate combined prediction is realized. The deep learning network can realize classification and extraction of space and time coupling characteristics in data and highlight important characteristics, and further improves the prediction accuracy of the model.
In order to make the above objects, features and advantages of the present invention more comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention.
FIG. 1 is a diagram of a joint prediction method technique route accounting for multiple uncertainties in the present invention;
FIG. 2 is an example data presentation diagram in accordance with the present invention;
FIGS. 3 (a) - (c) are graphs of cross-correlation analysis results in the present invention;
FIG. 4 is a graph of the results of an autocorrelation analysis in the present invention;
FIG. 5 is a schematic view of a feature extraction layer in the present invention;
FIG. 6 is a block diagram of a time series convolution attention module in accordance with the present invention;
FIG. 7 is a schematic diagram of a channel attention module of a time series convolution attention module of the present invention;
FIG. 8 is a timing attention module schematic of a timing convolution attention module of the present invention;
FIG. 9 is a schematic diagram of a feature sharing layer in the present invention;
FIG. 10 is a graph of a comparison of model predictions in the present invention;
FIG. 11 is a graph comparing MAPE probability distributions of model predictions in the present invention;
fig. 12 is a graph comparing RMSE probability distributions of model predictions in the present invention.
Detailed Description
The invention will be further described with reference to the drawings and examples.
It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the invention. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the present invention. As used herein, the singular is also intended to include the plural unless the context clearly indicates otherwise, and furthermore, it is to be understood that the terms "comprises" and/or "comprising" when used in this specification are taken to specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof.
As shown in FIG. 1, for multi-prediction targets and complex coupling characteristics thereof, in the longitudinal direction, the double-layer multi-task learning model combines commonalities and differences among different tasks, divides each task into three categories of a new energy prediction task, a multi-element load prediction task and a price prediction task, and divides a feature sharing model into an inner layer and an outer layer according to different task types in construction of a network model. And then, adopting a proper sharing mechanism among different levels to realize feature sharing among tasks under complex coupling characteristics, and enhancing the generalization capability of the model. In the transverse direction, the model adopts a CNN-SCAM-LSTM network, and firstly, the characteristic extraction of the whole input data is realized through CNN; then, adopting the proposed SCAM as an application of an attention mechanism, classifying the features according to different task types, and mining space-time features of the data to realize important feature extraction; and finally, constructing a feature sharing model through the LSTM, further acquiring important auxiliary coupling information according to the distinction between the inner layer and the outer layer, carrying out sharing learning in a hierarchical and targeted manner, and improving the model prediction precision.
It should be noted that, this example data is derived from wind power, electricity-heat load, electricity price and weather data in denmark 2010-2013 in the open source data set of the european electric power system. The influence of the installed capacity of the wind power on the wind power prediction data is considered, and the wind power data is adjusted according to different capacity ratios. The data of wind power, electricity-heat load and electricity price in 2010-2011 are shown in fig. 2.
And performing correlation analysis of the data.
Cross-correlation analysis
The complex energy system has complex relevance due to the mutual influence of multiple loads, new energy power generation, energy price, time and meteorological factors in the comprehensive energy system and the uncertainty of different coupling mechanisms and time sequences. In order to analyze the influence mechanism between each influence factor and the predicted object, relevant information in the multivariate data is fully mined, the input variables of the joint prediction model are effectively selected, a corresponding prediction network structure is constructed, and quantitative analysis is needed to be carried out on the data set.
The invention adopts the Pearson product difference correlation coefficient, the Spearman rank correlation coefficient and the Kendall rank correlation coefficient to discuss the correlation among all influencing factors and prediction targets.
Pearson product difference correlation coefficient
The Pearson correlation coefficient is used to measure the linear correlation between the two variables X and Y. It has a value between +1 and-1, where 1 is the total positive linear correlation, 0 is the nonlinear correlation, and-1 is the total negative linear correlation. One key mathematical property of the Pearson correlation coefficient is that it is invariant under individual changes in the position and scale of the two variables.
The calculation formula corresponding to the Pearson product difference correlation coefficient is as follows:
Figure BDA0004220501650000081
spearman rank correlation coefficient
The Spearman correlation coefficient uses the rank order of two variables for linear correlation analysis, and does not require the distribution of the original variables, and belongs to a non-parameter statistical method. Therefore, the application range of the method is far wider than that of the Pearson correlation coefficient. The Spearman correlation coefficient can be calculated even though the original data is the gradation data. Spearman correlation coefficients may also be calculated for Pearson correlation coefficient-compliant data, but with a somewhat lower statistical power than Pearson correlation coefficients. If there are no duplicate values in the data, and when the two variables are perfectly monotonically correlated, the spearman correlation coefficient is either +1 or-1.
For sample data with a sample capacity of n, a calculation formula corresponding to the Spearman rank correlation coefficient is as follows:
Figure BDA0004220501650000091
wherein d i For data X i And Y i Level differences between. The number of the columns is ranked from small to large, and if the numbers are the same, the positions of the columns are arithmetically averaged.
Kendall rank correlation coefficient
The Kendall correlation coefficient is also a rank correlation coefficient, and is used for reflecting the index of the correlation of the classified variables, and is suitable for the situation that two variables are classified in order, and the value of the Kendall correlation coefficient is represented by Greek letter tau.
The calculation formula corresponding to Kendall rank correlation coefficient is as follows:
Figure BDA0004220501650000092
wherein C represents the pair of element pairs (two elements are a pair) having consistency in the sample data X and Y; d represents the logarithm of the element in sample data X and Y that has inconsistencies.
In general, pearson correlation is a statistic for distance variables, spearman correlation is a statistic for sequencing variables, kendall correlation is a statistic for class variables, and example analysis results are shown in fig. 3 (a) - (c). The cross-correlation analysis results let us know how to choose the appropriate influencing factors as the input dataset for the predictive network. It can be seen that the correlation between each predicted target and the influencing factor is consistent under the measurement of different correlation coefficients. The correlation and the degree of correlation are that the electric load has medium correlation with the heat load and the electricity price and has weak correlation with the temperature and the irradiance; the thermal load shows stronger negative correlation with temperature, and the thermal-electric coupling influences the medium correlation with the electric load; the electricity price has medium correlation with the electric load, accords with the supply-demand relation between the load and the price, and has weak correlation with wind power at the power supply side; wind power and wind speed show strong correlation and are mainly influenced by weather factors.
Autocorrelation analysis
The cross-correlation analysis cannot intuitively embody the change rule of each prediction target at the time level, which causes the determination of the time sequence length of the input data to lack a theoretical basis. This faces the problem that if the time series is too long, feature redundancy can be created, allowing the model to learn many unnecessary parameters. Therefore, the calculation load of the model is increased, the prediction accuracy of the model on the test set is reduced, and the model is over-fitted. Too short to be sufficient for highly accurate predictions of highly non-linear time series. In this regard, analysis of the timing characteristics of the data may be performed using autocorrelation coefficients (Autocorrelation Function, ACF). The calculation formula of the autocorrelation coefficient is:
Figure BDA0004220501650000101
where k represents the delay time,
Figure BDA0004220501650000102
represents the mean value and n represents the total length of time.
Fig. 4 shows the data correlation after two weeks of electrical load, thermal load, electricity price and wind power delay. It can be seen that the electrical load is periodic every day and every week; the heat load is periodic every day; the electrical load then exhibits a daily correlation and a somewhat weaker weekly correlation; wind power has no obvious time characteristics. For the analysis results, 24h or 24 x 7h is selected as the data length.
And adopting a CNN network to extract the characteristics based on the correlation analysis result.
The feature extraction layer structure adopted by the invention is shown in fig. 5. First, feature extraction is performed on input data through a convolution layer, and the operation core is a convolution kernel. The convolution kernel regularly traverses the input data, performs matrix element multiplication summation on the input data in the receptive field, and superimposes the deviation amount. The feature map of the convolutional layer output is then passed to the pooling layer through activation of the ReLU function. The pooling layer belongs to a downsampling process and is mainly used for reducing data dimension and avoiding overfitting. Like the convolutional layer, the pooling layer operator consists of a fixed size window that also slides over all regions of the input according to the stride size, traversing each position through the window and computing the output. Finally, iterating according to the structure to obtain the final output characteristics.
The important feature extraction is further performed based on the attention module.
The attention mechanism model adopted by the invention is mainly improved by referring to a convolution attention module (Convolutional Block Attention Module, CBAM), and a time sequence convolution attention module (Sequential Convolution Attention Module, SCAM) is researched. As shown in FIG. 6, the SCAM model studied in the invention has an overall flow structure similar to that of CBAM and is formed by serially connecting a channel attention module (Channel Attention Module, CAM) and a time sequence attention module (Sequential Attention Module, SAM).
The model formula is expressed as:
Figure BDA0004220501650000111
Figure BDA0004220501650000112
in the formula, the input characteristic F epsilon R C*T Then the channel attention matrix M is obtained through the channel attention module c ∈R C*1 Matrix M c Obtaining the characteristics of the original imageSign F' ∈R C*T And is used as the input of the time sequence attention module, and the time attention matrix M is obtained through the time sequence attention module t ∈R C*T Then matrix M t Multiplying the original image to obtain output characteristic F'. Epsilon.R C*T 。
The overall structure of the channel attention module in the given SCAM model is unchanged from that in the CBAM, mainly the data dimension and the dimension of the hierarchy are adjusted, and the model structure is shown in fig. 7. Firstly, the input characteristic sequence passes through two parallel MaxPool layers and an AvgPool layer, and the dimension of the characteristic sequence is compressed from C to C1. Then, the channel number is compressed to 1/r times (Reduction rate) of the original channel number through a Share MLP module, and then the channel number is expanded back to the original channel number, and two activated results are obtained through a ReLU activation function. And finally, adding the two output results element by element, and obtaining a channel attention matrix through a sigmoid activation function.
The expression of the model is as follows:
Figure BDA0004220501650000121
the timing attention module in the SCAM model is shown in fig. 8. Firstly, splicing the data of different channels by adopting Concat operation according to the time dimension by the feature sequence processed by the channel attention module to obtain the feature sequence with dimension of 1 (C) and T. Then, the feature is extracted through a one-dimensional convolution layer with the convolution kernel size of 3, and padding is set to 3 so that the feature sequence dimension is unchanged. Finally, the dimension of the feature sequence is transformed into C.times.T by adopting view operation, and then the time sequence attention matrix is obtained through an activation function Sigmoid.
The expression of the model is as follows:
Figure BDA0004220501650000131
feature sharing is performed based on LSTM networks.
The data after the important feature extraction is transmitted to the feature sharing layer through the SCAM, the part mainly considers the correlation among prediction targets, and the information sharing is carried out on the features extracted by the pre-stage network, and the important point is to select a proper network structure and a proper feature sharing mechanism.
The source-charge-valence combined prediction model to be constructed by the invention has the characteristics of multiple parameters and complex structure, is not easy to generate over-fitting problem, and has stronger generalization capability. In the feature sharing part of the multi-task learning, the feature sharing model is divided into an inner layer and an outer layer according to different task types, and different sharing mechanisms are adopted. The feature sharing among three kinds of prediction tasks of source, load and price is taken as an outer layer, and a soft sharing mechanism is adopted in consideration of complex correlation and different influence mechanisms. Features of all subtasks in various tasks are shared as an inner layer, for example, load prediction is performed, coupling among multiple loads is strong, features of different loads are shared as an inner layer, and a hard sharing mechanism is adopted. In addition, considering the influence of the source and the load on the price, the price prediction is taken as a main task, the source and the load prediction are taken as auxiliary tasks to construct a double-layer feature sharing model, and the concrete model of the example is shown in fig. 9. The thermoelectric load is directly hard-shared through the LSTM network, wind power, load characteristics and electricity price characteristics are soft-shared among different LSTM network layers in a weight sum mode, and auxiliary tasks share information to a main task when the auxiliary tasks are output through a full-connection layer.
Example results analysis
In the invention, the construction and training of the source-charge-price double-layer combined prediction model are carried out under a PyTorch deep learning framework, an Intel Core i 7CPU is adopted by a hardware platform, and the calculation data are derived from the data in the year 2010-2012 of European Denmark. The training set and the verification set adopt 2010-2011 data, the test set adopts 2012 data, and the electricity-heat load, the wind power and the electricity price are predicted by taking 24h as step length.
The provided CNN-SCAM-LSTM-MTL model is compared with a single-task CNN-LSTM model, a multi-task CNN-LSTM-MTL model and a CNN-CBAM-LSTM-MTL model to verify the effectiveness of the designed double-layer multi-task combined prediction model.
As shown in fig. 10, the data from 1 month, 2 days, to 1 month, 8 days, and a week in the test set were selected for comparison of the prediction results. It can be seen that in the comparison of the single-task learning and the multi-task learning, the prediction effect of simply adopting the CNN-LSTM-MTL model is worse than that of the single-task learning, but the prediction results of the CNN-CBAM-LSTM-MTL and CNN-SCAM-LSTM-MTL models adopting the attention mechanism are different from those of the single-task learning CNN-LSTM model and are more similar to the actual values. This is because the data set adopted in the example has moderate correlation between prediction targets, and only the interaction of input features under the multi-task learning framework is adopted to put higher requirements on the link of important feature extraction. In this regard, the results verify the necessity to employ an attention mechanism.
The present example predicts annual data for a test set and calculates the daily prediction accuracy using the mean absolute percentage error (Mean Absolute Percentage Error, MAPE) and root mean square error (Root Mean Square Error, RMSE) as an evaluation index for the model, with the results shown in tables 1 and 2. It can be seen that the combined prediction is difficult to extract corresponding important features according to different prediction targets by adopting a multitask learning framework for the CNN-LSTM-MTL network, the prediction effect is not similar to that of a CNN-LSTM model, and the prediction effects of the CNN-CBAM-LSTM-MTL model and the CNN-SCAM-LSTM-MTL model are obviously improved after an attention mechanism is added. It follows that the attention mechanism employed is capable of extracting and classifying important features.
By comparing the CNN-LSTM model, the prediction accuracy of the CNN-CBAM-LSTM-MTL model on the electric-thermal load is still lower than that of single-task learning, but the wind power and electricity price prediction effect is good. This is because the electricity-heat load has a significant periodicity, and wind power and electricity prices are more random and fluctuating, and the time-series characteristics of the data are different from the time dimension. While CBAM is capable of extracting important features from the spatial dimension (i.e., channels in the feature data), it superimposes feature data for different channels from the temporal dimension to obtain a common attention weight matrix. This makes it impossible for the CBAM to extract timing feature differences in the instance data, for which the invention has been studied.
Compared with the CNN-LSTM model and the CNN-CBAM-LSTM-MTL model, the CNN-SCAM-LSTM-MTL model studied by the invention has better comprehensive prediction precision. The SCAM studied by the invention solves the defect of extracting the time sequence characteristics in the CBAM, better considers the space and time characteristics of the characteristic information, and can decouple the input characteristics according to different prediction targets from two dimensions of space and time, thereby improving the prediction precision. In this regard, the effectiveness of the source-charge-price double-layer joint prediction model and SCAM based on the multi-task learning, which are researched by the invention, can be verified.
TABLE 1MAPE index comparison
Figure BDA0004220501650000161
TABLE 2RMSE index comparison
Figure BDA0004220501650000162
In this example, probability distribution information of prediction accuracy under MAPE and RMSE indexes in 2012 was counted in days, the probability density functions thereof are shown in FIG. 11 and FIG. 12, and the confidence intervals with the confidence of 90% are shown in tables 3 and 4. In the whole, the probability peak value and the confidence interval of the prediction precision of the CNN-SCAM-LSTM-MTL network are better, the stability of the prediction result is reflected, the multi-task learning model studied by the invention is further verified to have better prediction precision, and the generalization of the model can be improved by weighing training information in the multi-task learning model.
TABLE 3MAPE indicator confidence interval comparison
Figure BDA0004220501650000163
Figure BDA0004220501650000171
TABLE 4 confidence interval comparison of RMSE indicators
Figure BDA0004220501650000172
It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
While the foregoing description of the embodiments of the present invention has been presented in conjunction with the drawings, it should be understood that it is not intended to limit the scope of the invention, but rather, it is intended to cover all modifications or variations within the scope of the invention as defined by the claims of the present invention.

Claims (10)

1. The comprehensive energy system source charge price multitasking joint prediction method is characterized by comprising the following steps:
analyzing the space-time coupling characteristics among all uncertainty factors affecting the comprehensive energy system from the two angles of cross correlation and autocorrelation;
extracting features from the two correlation analysis results;
extracting the extracted features for the second time by using a channel attention and time sequence attention mechanism;
and carrying out feature sharing on the data after secondary extraction by using a parameter sharing learning mechanism, wherein the multitask learning method takes price prediction as a main task, source and load prediction as auxiliary tasks, coupling among corresponding loads of various tasks adopts a hard sharing mechanism, and soft sharing mechanisms are adopted among different types of tasks, so that the auxiliary tasks share information to the main task, and a joint prediction result is obtained.
2. The method for combined prediction of the source charge price of the comprehensive energy system according to claim 1, wherein the method is characterized in that the correlation analysis among the multi-element load, the new energy power generation, the energy price, the time and the meteorological factors is respectively carried out by utilizing the Pearson product difference correlation coefficient, the Spearman rank correlation coefficient and the Kendall rank correlation coefficient.
3. The method for combined prediction of source charge and price in an integrated energy system according to claim 1, wherein the time sequence characteristics of each factor are analyzed by using autocorrelation coefficients to determine the data time length adopted for prediction.
4. The method for combined prediction of source charge price and multiple tasks of an integrated energy system according to claim 1, wherein the specific process of feature extraction of two correlation analysis results comprises: extracting features of input data through a convolution layer, wherein the convolution layer regularly traverses the input data, performs matrix element multiplication summation on the input data and superimposes deviation values;
transmitting the feature map output by the convolution layer to a pooling layer by utilizing activation of a ReLU function, sliding a window of the pooling layer on all areas of the input according to the stride size, traversing each position through the window, and calculating the output;
and (5) iterating the process to obtain the final output characteristic.
5. The method for combined prediction of source charge and price in an integrated energy system according to claim 1, wherein the specific process of extracting the extracted features for the second time by using a channel attention and time sequence attention mechanism comprises the following steps: the channel attention module is connected with the time sequence attention module in series, the input characteristics are used for obtaining a channel attention matrix through the channel attention module, the matrix is multiplied by the original image to obtain characteristics and used as the input of the time sequence attention module, the time attention matrix is obtained through the time sequence attention module, and then the time attention matrix is multiplied by the original image to obtain output characteristics.
6. The method for combined prediction of source charge and price of a comprehensive energy system according to claim 1 or 5, wherein the processing procedure of the channel attention module comprises the steps of compressing the dimension of an input characteristic sequence through two parallel MaxPool layers and an AvgPool layer, compressing the number of channels, expanding the number of channels back to the number of original channels, and obtaining two activated results through a ReLU activation function; adding the two output results element by element, and obtaining a channel attention matrix through a sigmoid activation function;
or alternatively, the first and second heat exchangers may be,
the processing procedure of the time sequence attention module comprises the following steps: splicing the data of different channels according to the time dimension by the feature sequence processed by the channel attention module to obtain the feature sequence; and extracting features by a one-dimensional convolution layer, converting the dimension of the feature sequence into the original dimension, and obtaining a time sequence attention matrix through an activation function Sigmoid.
7. The method for predicting the combined charge and price of the comprehensive energy system according to claim 1 is characterized in that the LSTM network is utilized for feature sharing, the thermoelectric load is directly hard-shared through the LSTM network, wind power, load features and electricity price features are soft-shared among different LSTM network layers in a weight sum mode, and auxiliary tasks share information to main tasks during output.
8. The utility model provides a comprehensive energy system source charge price multitasking joint prediction system which characterized in that includes:
a correlation analysis module configured to analyze a spatiotemporal coupling characteristic between uncertainty factors affecting the integrated energy system from both a cross correlation and an autocorrelation perspective;
the feature extraction module is configured to perform feature extraction on two correlation analysis results;
the secondary extraction module is configured to perform secondary extraction on the extracted features by using a channel attention and time sequence attention mechanism;
the feature sharing module is configured to construct a double-layer feature sharing model by utilizing a hard sharing mechanism and a soft sharing mechanism, and perform feature sharing on the data after secondary extraction, wherein the double-layer feature sharing model adopts the hard sharing mechanism according to the division of the predicted task types, the coupling among the corresponding loads of various tasks adopts the hard sharing mechanism, and adopts the soft sharing mechanism among different types of tasks, so that auxiliary tasks share information to main tasks, and a joint prediction result is obtained.
9. A computer readable storage medium, characterized in that a plurality of instructions are stored, said instructions being adapted to be loaded by a processor of a terminal device and to perform the steps of the method of any of claims 1-7.
10. A terminal device, comprising a processor and a computer readable storage medium, the processor configured to implement instructions; a computer readable storage medium for storing a plurality of instructions adapted to be loaded by a processor and to perform the steps of the method of any of claims 1-7.
CN202310519941.5A 2023-05-06 2023-05-06 Comprehensive energy system source charge price multitasking combined prediction method and system Pending CN116432861A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310519941.5A CN116432861A (en) 2023-05-06 2023-05-06 Comprehensive energy system source charge price multitasking combined prediction method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310519941.5A CN116432861A (en) 2023-05-06 2023-05-06 Comprehensive energy system source charge price multitasking combined prediction method and system

Publications (1)

Publication Number Publication Date
CN116432861A true CN116432861A (en) 2023-07-14

Family

ID=87089132

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310519941.5A Pending CN116432861A (en) 2023-05-06 2023-05-06 Comprehensive energy system source charge price multitasking combined prediction method and system

Country Status (1)

Country Link
CN (1) CN116432861A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117132135A (en) * 2023-10-23 2023-11-28 陕西天润科技股份有限公司 Urban informatization management system and method based on digital twinning
CN117369282A (en) * 2023-11-17 2024-01-09 上海四方无锡锅炉工程有限公司 Control method for adaptive hierarchical air supply and solid waste CFB boiler thereof

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115471362A (en) * 2022-09-26 2022-12-13 东南大学 Comprehensive energy source-load prediction method for depth feature-guided two-stage transfer learning
CN115630278A (en) * 2022-10-27 2023-01-20 重庆大学 Network vibration damage detection method based on channel-space-time attention mechanism
CN115759458A (en) * 2022-12-01 2023-03-07 国网天津市电力公司 Load prediction method based on comprehensive energy data processing and multi-task deep learning
CN115829115A (en) * 2022-11-28 2023-03-21 浙江工业大学 PCA-LSTM-MTL-based photovoltaic power station-containing area load prediction method
US20230095676A1 (en) * 2021-09-27 2023-03-30 Hohai University Method for multi-task-based predicting massiveuser loads based on multi-channel convolutional neural network
CN115936218A (en) * 2022-12-06 2023-04-07 山东大学 Comprehensive energy system multi-element load prediction method and system based on deep learning

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20230095676A1 (en) * 2021-09-27 2023-03-30 Hohai University Method for multi-task-based predicting massiveuser loads based on multi-channel convolutional neural network
CN115471362A (en) * 2022-09-26 2022-12-13 东南大学 Comprehensive energy source-load prediction method for depth feature-guided two-stage transfer learning
CN115630278A (en) * 2022-10-27 2023-01-20 重庆大学 Network vibration damage detection method based on channel-space-time attention mechanism
CN115829115A (en) * 2022-11-28 2023-03-21 浙江工业大学 PCA-LSTM-MTL-based photovoltaic power station-containing area load prediction method
CN115759458A (en) * 2022-12-01 2023-03-07 国网天津市电力公司 Load prediction method based on comprehensive energy data processing and multi-task deep learning
CN115936218A (en) * 2022-12-06 2023-04-07 山东大学 Comprehensive energy system multi-element load prediction method and system based on deep learning

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
ZHICHAO ZHANG，ET AL.: ""Learning Attentive Representations for Environmental Sound Classification"", 《IEEE ACCESS》, vol. 7, 24 September 2019 (2019-09-24) *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117132135A (en) * 2023-10-23 2023-11-28 陕西天润科技股份有限公司 Urban informatization management system and method based on digital twinning
CN117132135B (en) * 2023-10-23 2024-02-06 陕西天润科技股份有限公司 Urban informatization management system and method based on digital twinning
CN117369282A (en) * 2023-11-17 2024-01-09 上海四方无锡锅炉工程有限公司 Control method for adaptive hierarchical air supply and solid waste CFB boiler thereof
CN117369282B (en) * 2023-11-17 2024-04-19 上海四方无锡锅炉工程有限公司 Control method for adaptive hierarchical air supply and solid waste CFB boiler thereof

Similar Documents

Publication Publication Date Title
Cannizzaro et al. Solar radiation forecasting based on convolutional neural network and ensemble learning
Baker et al. Evaluating lossy data compression on climate simulation data within a large ensemble
Ozer et al. A combined deep learning application for short term load forecasting
CN116432861A (en) Comprehensive energy system source charge price multitasking combined prediction method and system
CN111091196B (en) Passenger flow data determination method and device, computer equipment and storage medium
Ozkan et al. Data mining-based upscaling approach for regional wind power forecasting: Regional statistical hybrid wind power forecast technique (RegionalSHWIP)
Wang et al. Comparative performance of logistic regression and survival analysis for detecting spatial predictors of land-use change
Al-Janabi et al. Development of deep learning method for predicting DC power based on renewable solar energy and multi-parameters function
Lenzi et al. Benefits of spatiotemporal modeling for short‐term wind power forecasting at both individual and aggregated levels
Jove et al. Missing data imputation over academic records of electrical engineering students
CN113591368A (en) Comprehensive energy system multi-energy load prediction method and system
Oprea A general framework and guidelines for benchmarking computational intelligence algorithms applied to forecasting problems derived from an application domain-oriented survey
Pravallika et al. Prediction of temperature anomaly in Indian Ocean based on autoregressive long short-term memory neural network
Mohammad et al. Short term load forecasting using deep neural networks
Frimane et al. Infinite hidden Markov model for short-term solar irradiance forecasting
Somu et al. Evaluation of building energy demand forecast models using multi-attribute decision making approach
Madhavi et al. Multivariate deep causal network for time series forecasting in interdependent networks
Jalalifar et al. SAC-ConvLSTM: A novel spatio-temporal deep learning-based approach for a short term power load forecasting
Giebel et al. Simulation and prediction of wind speeds: A neural network for Weibull
Karozis et al. A deep learning approach for spatial error correction of numerical seasonal weather prediction simulation data
Jiang et al. Evolution characteristics of potential evapotranspiration over the Three-River Headwaters Region
Chen et al. Explainable Global Wildfire Prediction Models using Graph Neural Networks
El Bakali et al. Solar radiation forecasting using artificial intelligence techniques for energy management system
Roh et al. Tfe-net: time and feature focus embedding network for multivariate-to-multivariate time series forecasting
Xu et al. Calibration and sensitivity analysis of long-term generation investment models using Bayesian emulation

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