CN116845889B - Hierarchical hypergraph neural network-based power load prediction method - Google Patents
Hierarchical hypergraph neural network-based power load prediction method Download PDFInfo
- Publication number
- CN116845889B CN116845889B CN202311123122.5A CN202311123122A CN116845889B CN 116845889 B CN116845889 B CN 116845889B CN 202311123122 A CN202311123122 A CN 202311123122A CN 116845889 B CN116845889 B CN 116845889B
- Authority
- CN
- China
- Prior art keywords
- hypergraph
- scale
- nodes
- time
- node
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 40
- 230000003993 interaction Effects 0.000 claims abstract description 38
- 230000007246 mechanism Effects 0.000 claims abstract description 19
- 230000004931 aggregating effect Effects 0.000 claims abstract description 9
- 238000013507 mapping Methods 0.000 claims abstract description 9
- 230000005540 biological transmission Effects 0.000 claims abstract description 8
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 239000011159 matrix material Substances 0.000 claims description 51
- 238000010276 construction Methods 0.000 claims description 19
- 238000000605 extraction Methods 0.000 claims description 14
- 230000002776 aggregation Effects 0.000 claims description 12
- 238000004220 aggregation Methods 0.000 claims description 12
- 230000002123 temporal effect Effects 0.000 claims description 10
- 230000006870 function Effects 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 4
- 238000012546 transfer Methods 0.000 claims description 4
- 235000008694 Humulus lupulus Nutrition 0.000 claims description 3
- 238000007726 management method Methods 0.000 abstract description 5
- 238000013439 planning Methods 0.000 abstract description 5
- 230000008859 change Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 3
- 230000005611 electricity Effects 0.000 description 3
- 230000007774 longterm Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000009191 jumping Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013450 outlier detection Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000006116 polymerization reaction Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000001932 seasonal effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012384 transportation and delivery Methods 0.000 description 1
- 238000002759 z-score normalization Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- 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
- G06N3/042—Knowledge-based neural networks; Logical representations of neural 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
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
-
- 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)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Business, Economics & Management (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Economics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Public Health (AREA)
- Power Engineering (AREA)
- Water Supply & Treatment (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an electric load prediction method based on a hierarchical hypergraph neural network, which comprises the following steps of: preprocessing the electric load data and constructing a training sample; mapping the sequence of training samples into subsequences of different time scales; constructing a hierarchical hypergraph according to a time proximity rule to model high-order interactions among scales, inter-scales and mixed-scale time modes, and constructing a hyperedge graph based on a sequential relationship and an association relationship among nodes; aggregating time mode information of different scales through a node-superside, superside-superside and a three-stage message transmission mechanism among the superside-nodes, and learning interaction strength among the time modes of different scales; fusing time mode information of different scales, and training an electric load prediction model based on all training samples; and (5) carrying out electric load prediction. The method can improve the accuracy of electric load prediction, and is suitable for application scenes such as electric power system operation, energy planning, energy efficiency management and the like.
Description
Technical Field
The invention belongs to the technical field of electric load prediction, and particularly relates to an electric load prediction method based on a hierarchical hypergraph neural network.
Background
With the continuous promotion of urban and industrialized production, the electric power demand is continuously increased, and the gradual increase of urban electric loads becomes an important problem. This situation not only affects people's daily life and work, but also presents significant challenges for related departments to power planning and energy management. The method can be used for accurately predicting the change of the future power load, effectively guiding people to supply power and allocate energy, and simultaneously providing powerful data support for power system management. Therefore, electrical load prediction is one of the very important and valuable research directions in the energy field.
The electric load data in real life exhibits a complex and diverse time pattern such as a day pattern (early peak and late peak) and a week pattern (in the week and weekend) due to the influence of the periodic human activities. Considering interactions between these temporal patterns can generally achieve more accurate predictions than analyzing each pattern alone, and is of great importance for both long-term planning and early warning of electrical loads.
In order to model time patterns of different scales and interactions thereof in power load data, traditional methods such as Seasonal autoregressive moving average (seasal ARIMA) and Prophet model (Prophet) use decomposition methods based on heuristic prior knowledge to acquire the time patterns of different scales, but cannot model nonlinear dependency of time series. To capture non-stationary and non-linear time dependencies, time convolutional networks (Temporal Convolutional Networks, TCNs), recurrent neural networks (Recurrent Neural Networks, RNNs), and Transformers are used for power load prediction.
To model temporal patterns of different scales, the hierarchical Transformer method attempts to construct subsequences of different scales from the original electrical load sequence. The Chinese patent application with publication number of CN114529051A discloses a long-term power load prediction method based on a hierarchical residual self-attention neural network, which effectively captures the long-term and short-term characteristics of a sequence by extracting the mixed characteristics of trend items, period items, holiday items and weather items in historical load data and hierarchically dismantling, reconstructing and predicting the load sequence. But this approach ignores interactions between different scale temporal patterns.
Some hierarchical transform methods introduce special structures such as pyramids among sub-sequences of different scales, model time dependency relationships in the sub-sequences by edges among nodes of the same scale, and model interactions among time modes of different scales by edges among nodes of different scales. The Chinese patent application with publication number of CN113988415A discloses a medium-long term power load prediction method, which comprises the steps of constructing CNN-LSTM network of a fused feature pyramid to predict nonlinear components of high frequency, constructing ARIMA model to predict linear components of low frequency, and separately predicting and correcting loads of season inflection points and months.
In fact, there is a high-order interaction of the time patterns of multiple scales in the power load sequence at the same time, for example, the peak of the family power consumption in the weekend in summer appears in the afternoon (the high-order interaction of the day, week and month patterns), and the family power consumption in the weekend in winter is high and has small fluctuation. Thus, while the above approach achieves a certain effect, it lacks the ability to model higher order interactions between different scale temporal patterns in a power load sequence.
Disclosure of Invention
In view of the above, an object of the present invention is to provide a power load prediction method based on a hierarchical hypergraph neural network, which performs hierarchical feature extraction on power load data, constructs hypergraphs and hyperedge graphs to explicitly represent high-order interactions between time modes of different scales, and introduces a three-stage message passing mechanism to aggregate time mode information of different scales and learn interaction strength thereof, so as to fully discover and utilize the high-order interaction information between the time modes of different scales, and is suitable for application scenarios such as power system operation, energy planning, energy efficiency management, and the like.
In order to achieve the above purpose, the technical scheme provided by the invention is as follows:
the embodiment of the invention provides a power load prediction method based on a hierarchical hypergraph neural network, which comprises the following steps:
s1, preprocessing electric load data and constructing a training sample;
s2, mapping the sequence of the training sample into subsequences with different time scales by using a layering characteristic extraction module;
s3, in a hypergraph-hyperedge graph construction module, electric load data at different moments in a subsequence are regarded as nodes, and a hierarchical hypergraph is constructed according to a time proximity rule so as to model high-order interactions among intra-scale, inter-scale and mixed-scale time modes; constructing a hypergraph based on the sequence relation and the association relation among the nodes by taking the hyperedge of the hypergraph as the node;
s4, in the three-stage message transfer module, the time mode information of different scales is aggregated through a node-superside, a superside-superside and a three-stage message transfer mechanism among the superside-nodes, and the interaction strength among the time modes of different scales is learned;
s5, fusing time mode information of different scales, carrying out electric load prediction, and training an electric load prediction model based on all training samples, wherein the electric load prediction model comprises a layering characteristic extraction module, a hypergraph-hyperedge graph construction module and a three-stage message transmission module;
s6, inputting the electric load sequence to be predicted into a trained electric load prediction model to obtain a prediction result.
Preferably, in S1, the preprocessing the electrical load data and constructing a training sample includes: preprocessing given electric load data, including outlier processing and normalization processing; dividing the preprocessed power load data through a sliding time window to obtain a training data set; the training data set is batched according to the fixed batch size, and training samples are selected according to batches.
Preferably, in S2, the mapping the sequence of training samples into subsequences with different time scales by using the hierarchical feature extraction module includes: the sequence of the training sample is input into a layering characteristic extraction module, aggregation functions are utilized to generate subsequences with different time scales by setting aggregation windows with different sizes, the length of the next time scale subsequence is the length of the last time scale subsequence divided by the size of the aggregation window corresponding to the last time scale, and finally all the subsequences form a subsequence set.
Preferably, in S3, the step of regarding the power load data at different moments in the subsequence as nodes, and constructing a hierarchical hypergraph according to a time proximity rule to model high-order interactions between intra-scale, inter-scale and mixed-scale time modes includes: taking the power load data at different moments in the subsequence as nodes, constructing an original intra-scale hypergraph, an original inter-scale hypergraph and an original mixed-scale hypergraph according to a time proximity rule, and respectively constructing hypergraphs of each typekJump connection to aggregate information in different neighbor rangeskJump-scale hypergraph,kInter-scale hopping hypergraphkJump hybrid scale hypergraph, whereinkFor the time span between two adjacent nodes, finally obtaining the graph containing the original intra-scale hypergraph andkan intra-scale hypergraph that jumps to an intra-scale hypergraph, comprising an original inter-scale hypergraph andkinter-scale hypergraphs that skip inter-scale hypergraphs, comprising an original mixed-scale hypergraph andka hybrid scale hypergraph that hops the hybrid scale hypergraph.
Preferably, the intra-scale hypergraph is formed by constructing an original intra-scale hypergraph by constructing original connections between adjacent nodes in the same scale according to correlation to obtain multiple hyperedges, and constructing nodes with time span and correlation information in the same scalekJump connection to obtain multiple superedge constructionkJumping the hypergraph in the scale, and summing the hypergraph in the original scalekAnd obtaining the intra-scale hypergraph by splicing the intra-scale hypergraph.
Preferably, the inter-scale hypergraph is obtained by constructing original connection between adjacent nodes in the same scale and nodes in the next scale according to correlation to obtain multiple hyperedges, and constructing the original inter-scale hypergraph with the same dimensionEstablishment between node with time span and correlation information in scale and next scale nodekJump connection to obtain multiple superedge constructionkJumping the hypergraph between scales, summing the hypergraph between original scaleskAnd obtaining the inter-scale hypergraph by splicing the inter-scale hypergraph.
Preferably, the mixed scale hypergraph is formed by constructing original connection of nodes of each scale according to correlation to obtain multiple hyperedges, wherein nodes in the same scale connected by the hyperedges are adjacent nodes, and the nodes of each scale are constructedkJump connection to obtain multiple superedge constructionkJumping mixed scale hypergraph, wherein nodes in the same scale connected by hyperedges are nodes with time span and correlation information, and combining the original mixed scale hypergraph withkAnd obtaining the mixed scale hypergraph by splicing the mixed scale hypergraph.
Preferably, in S3, the constructing the hypergraph by regarding the hyperedge of the hypergraph as a node based on the sequential relationship and the association relationship between the nodes includes:
in the intra-scale hypergraph, two hyperedges with time sequence nodes are connected to form a sequence relation hyperedge graph comprising node sets, edge sets and an adjacent matrix based on the sequence relation, wherein the hyperedges of the intra-scale hypergraph are regarded as nodes of the sequence relation hyperedge graph to form the node sets, the nodes of the sequence relation hyperedge graph form the edge sets according to the correlation, the nodes of the sequence relation hyperedge graph are regarded as the horizontal and vertical coordinates of the adjacent matrix, the value of 0 in the adjacent matrix indicates that the two hyperedges in the intra-scale hypergraph have no sequence relation, and the value of 1 in the adjacent matrix indicates that the two hyperedges in the intra-scale hypergraph have the sequence relation;
in the inter-scale hypergraph or the mixed-scale hypergraph, two hyperedges with common nodes have an association relationship, an association relationship hypergraph comprising node sets, edge sets and an adjacent matrix is constructed based on the association relationship, wherein the hyperedges of the inter-scale hypergraph or the mixed-scale hypergraph are regarded as nodes of the association relationship hypergraph to form the node sets, the nodes of the association relationship hypergraph form the edge sets according to the correlation construction edges, the nodes of the association relationship hypergraph are regarded as horizontal coordinates and vertical coordinates of the adjacent matrix, a value of 0 in the adjacent matrix indicates that the association relationship exists between the two hyperedges in the inter-scale hypergraph or the mixed-scale hypergraph, and a value of 1 in the adjacent matrix indicates that the association relationship does not exist between the two hyperedges in the inter-scale hypergraph or the mixed-scale hypergraph;
and filling and stacking the adjacency matrix constructed based on the sequential relation and the adjacency matrix constructed based on the association relation to obtain a total adjacency matrix, wherein the dimension of the total adjacency matrix is the sum of the nodes of the sequential relation superside graph and the nodes of the association relation superside graph.
Preferably, in S4, in the three-phase message passing module, the aggregating time mode information of different scales and learning interaction strength between time modes of different scales by a three-phase message passing mechanism of node-superside, superside-superside and superside-node includes:
s41, mapping the subsequence into an initialized node embedding, and obtaining the initialized superside embedding by aggregating the node embedding connected with the superside;
s42, updating the superside embedding through an attention mechanism based on the constructed superside graph, and outputting the updated superside embedding;
s43, dynamically updating the point-edge association matrix through a hypergraph attention mechanism based on the constructed hierarchical hypergraph, updating node embedding through hypergraph convolution, and outputting updated node embedding.
Preferably, in S6, the time mode information of different scales is fused, electric load prediction is performed, after the updated node is embedded, the last node of the subsequence of different scales is embedded and spliced to be input into a linear layer for fusion, and finally, an electric load predicted value is output.
Compared with the prior art, the method has the advantages that at least the following steps are included:
(1) According to the method, the hypergraph is built according to the time proximity rule, a basis is provided for modeling high-order interactions among the electric load data in-scale, inter-scale and mixed-scale time modes, and the hypergraph is built according to the interaction modes among the hyperedges by taking the hyperedges as nodes, so that the capability of modeling the high-order interactions among the time modes by the model is enhanced.
(2) The method adopts a three-stage message transmission mechanism to aggregate time mode information of different scales in the power load sequence and automatically learn the interaction strength among the time modes of different scales, thereby further improving the expression capacity of the model.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a general flow chart of a hierarchical hypergraph neural network-based power load prediction method provided by an embodiment of the invention;
FIG. 2 is a general framework diagram of a hierarchical hypergraph neural network-based power load prediction method provided by an embodiment of the invention;
FIG. 3 is a schematic diagram of hypergraph construction of a hierarchical hypergraph neural network-based power load prediction method according to an embodiment of the present invention;
fig. 4 is a schematic diagram of supergraph construction of a hierarchical supergraph neural network-based power load prediction method according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the detailed description is presented by way of example only and is not intended to limit the scope of the invention.
The invention is characterized in that: aiming at the problem that the electric load prediction method in the prior art cannot fully find and utilize high-order interaction among time modes of different scales, the embodiment of the invention provides the electric load prediction method based on the hierarchical hypergraph neural network, and the more accurate electric load prediction is realized by constructing hierarchical hypergraphs and hyperedge graphs to model high-order interaction among the time modes of different scales, introducing a three-stage message transmission mechanism to aggregate the time mode information of different scales and learn the interaction intensity.
Fig. 1 is an overall flowchart of an electrical load prediction method based on a hierarchical hypergraph neural network according to an embodiment of the present invention, and fig. 2 is an overall frame diagram of an electrical load prediction method based on a hierarchical hypergraph neural network according to an embodiment of the present invention. As shown in fig. 1 and 2, an embodiment provides a hierarchical hypergraph neural network-based power load prediction method, which includes the following steps:
defining electrical load prediction tasks as: before givingObservations of individual time stepsPredicting future->The value of the individual time steps。
And step 1, constructing an electric load time sequence according to the electricity consumption data acquired by the electric load sensor, and dividing the processed data through a sliding time window to obtain a training data set.
Specifically, preprocessing given electrical load data, including outlier processing and normalization processing; and dividing the preprocessed electric load data through a sliding time window to obtain a training data set.
In an embodiment, the sequence of given electrical load data is subjected to outlier detection and invalid values therein (e.g., values outside the normal range and zero values) are removed. The data with outlier removal is subjected to Z-score normalization for converting the data into a standard normal distribution with zero mean and unit variance, and the conversion formula is as follows:
（1）
wherein,the time step in the power load sequence after eliminating the abnormal value is +>Numerical value of>For the average value of the electrical load sequence data, +.>For the standard deviation of the electrical load sequence, +.>For the time step->After normalization of the data of (a).
Artificially setting the size of a time window according to experienceAnd dividing the normalized data by utilizing the sliding step length with fixed length to obtain a training data set.
Step 2, the training data set is batched according to the fixed batch size, and the total batch number is。
Specifically, the training data set is batched according to a fixed batch size, and training samples are selected according to batches.
In an embodiment, the batch is set manually based on experienceSize of the amountBatching training data sets with total batch count. The specific calculation mode is as follows:
（2）
wherein,is the total number of samples in the training dataset.
Step 3, sequentially selecting indexes from the training data set asIs a training sample of (1), wherein->. Steps 4-9 are repeated for each training sample in the batch.
Step 4, training sample sequenceInputting the extracted signals into a hierarchical feature extraction module to construct +.>Sub-sequence set of individual scales->. Wherein->For the original input sample sequence, +.>Is->Sub-sequences of individual scale,/->The value range is +.>。
Specifically, the sequence of the training sample is input into a layering characteristic extraction module, aggregation functions are utilized to generate subsequences with different time scales by setting aggregation windows with different sizes, the length of the next time scale subsequence is the length of the last time scale subsequence divided by the size of the aggregation window corresponding to the last time scale, and finally all the subsequences form a subsequence set.
In an embodiment, as shown in FIG. 2 (a), a sample power load sequence is givenThe hierarchical feature extraction module generates +_by setting different aggregation windows>A sub-sequence of the individual scale->The subsequences of the individual scales are expressed asWherein->Is->Personal dimension->The sequence of electrical loads at the moment in time is indicative,is->Sequence length of individual scale, +.>Is->Aggregate window size of individual dimension, +.>Is the characteristic dimension of the sequence. The specific polymerization process can be expressed as:
（3）
wherein,for aggregation functions, e.g. convolution or pooling, etc, +.>In the first place as an aggregate functionA learnable parameter of individual scale,/->For real number set->Is->Sequence length of individual scales.
Finally, the hierarchical feature extraction module generatesSub-sequence set of individual scales->。
Step 5, for the constructed sub-sequence setConstructing three types of hypergraph structures including intra-scale, inter-scale and mixed-scale according to a time proximity rule, and finally generating a final hypergraph according to the three types of hypergraph structures>Wherein->And->Representing a node set and a superside set, respectively.
Specifically, the power load data at different moments in the subsequence are regarded as nodes, an original intra-scale hypergraph, an original inter-scale hypergraph and an original mixed-scale hypergraph are constructed according to a time proximity rule, and hypergraphs of each type are respectively builtkJump connection to aggregate information in different neighbor rangeskJump-scale hypergraph,kInter-scale hopping hypergraphkJump hybrid scale hypergraph, whereinkFor the time span between two adjacent nodes, finally obtaining the graph containing the original intra-scale hypergraph andkan intra-scale hypergraph that jumps to an intra-scale hypergraph, comprising an original inter-scale hypergraph andkinter-scale hypergraphs that skip inter-scale hypergraphs, comprising an original mixed-scale hypergraph andka hybrid scale hypergraph that hops the hybrid scale hypergraph.
In an embodiment, a hypergraph construction schematic is shown in FIG. 3, which models high-order interactions between intra-scale, inter-scale, and mixed-scale temporal patterns by constructing different types of hypergraph structures on the one hand, and creates for each type of hypergraph on the other handkThe hop connection aggregates information in different neighbor ranges.
(5-1) constructing an intra-scale hypergraph, constructing original connection between adjacent nodes in the same scale according to correlation to obtain a plurality of hyperedges, constructing the original intra-scale hypergraph, and constructing nodes with time span and correlation information in the same scalekJump connection to obtain multiple superedge constructionkJumping the hypergraph in the scale, and summing the hypergraph in the original scalekThe jump-scale hypergraph is obtained through splicing operationIntra-scale hypergraph.
In an embodiment, as shown in FIG. 3 (a), to model high-order interactions between intra-scale temporal patterns, an intra-scale hypergraph is constructedWhich comprises an original intra-scale hypergraph +.>Andkjump scale inner hypergraph，/>And->Respectively, different-scale original intra-scale hyperedge set sumkAnd (5) a superset set in the jump scale.
Based on the first connectionThe (th) of the individual scale>Strip superb->The definition is as follows:
（4）
wherein,is->The individual scale start index is +.>Node of->Is->The individual scale index is +.>Is provided with a node which is connected with a node,is->Is->Is based on the first part of the original connection>The (th) of the individual scale>Start index of bar superside connection node, +.>Index for arbitrary node->Is->The number of nodes connected by the hyperedge of the individual scale,is a constraint.
Based onJump connection->The (th) of the individual scale>Strip superb->The definition is as follows:
（5）
wherein,is based onkTime span of two adjacent nodes of the hop connection, < >>Is->The individual scale start index is +.>Node of->The starting index of the node under each superside can be calculated by the following formula:
（6）
based on original connection hypergraphkIntra-scale hypergraph of jump connection hypergraphThe definition is as follows:
（7）
wherein,is a splicing operation.
Higher order interactions of the temporal patterns exist not only within the same scale, but also between adjacent scales. For example, a higher order interaction between hour and day modes. In addition, there are also high-order interactions between mixed-scale temporal patterns, such as between hour, day, week and month patterns. Therefore, the embodiment of the invention also constructs inter-scale hypergrams and mixed-scale hypergrams.
(5-2) constructing an inter-scale hypergraph, constructing original connection between adjacent nodes in the same scale and nodes in the next scale according to correlation to obtain a plurality of hyperedges, constructing the original inter-scale hypergraph, and constructing a node which has time span and correlation information in the same scale and the nodes in the next scalekJump connection to obtain multiple superedge constructionkJumping the hypergraph between scales, summing the hypergraph between original scaleskAnd obtaining the inter-scale hypergraph by splicing the inter-scale hypergraph.
In an embodiment, as shown in FIG. 3 (b), an inter-scale hypergraphContains the original inter-scale hypergraphAndkjump-scale hypergraph->，/>Andrespectively, different scales of the original inter-scale hyperedge set sumkAnd (5) a super-edge set among the jump scales.
Based on the first connectionThe (th) of the individual scale>Strip superb->The definition is as follows:
（8）
wherein,is->Aggregate windows of individual dimensions.
Based onJump connection->The (th) of the individual scale>Strip superb->The definition is as follows:
（9）
final inter-scale hypergraphThe definition is as follows:
（10）。
(5-3) constructing a mixed scale hypergraph, namely constructing original connection of nodes of all scales according to the correlation to obtain a plurality of hyperedges, constructing the original mixed scale hypergraph, wherein nodes in the same scale connected by the hyperedges are adjacent nodes, and constructing the nodes of all scaleskJump connection to obtain multiple superedge constructionkJumping mixed scale hypergraph, wherein nodes in the same scale connected by hyperedges are nodes with time span and correlation information, and combining the original mixed scale hypergraph withkThe jump mixed scale hypergraph is obtained through splicing operationHybrid scale hypergraph.
In an embodiment, as shown in FIG. 3 (c), a hybrid scale hypergraphComprises the original mixed scale hypergraph->Andkjump hybrid scale hypergraph->，/>And->Respectively, the original mixed scale superedge sets andkand (5) jumping the mixed scale superlimit set.
Based on the first connectionStrip superb->The definition is as follows:
（11)
wherein,is->The individual scale index is +.>Node of->In order to be node under each superside +.>Start index of the individual scale.
Based onkJump connection firstStrip superb->The definition is as follows:
(12)
final mixed-scale hypergraphThe definition is as follows:
(13)
hypergrams generated by splicing based on intra-scale hypergrams, inter-scale hypergrams and mixed-scale hypergramsThe definition is as follows:
(14)。
step 6, hypergraph is obtainedThe supersides of (a) are regarded as nodes, and interaction modes of two sides are defined: sequential relationship and association relationship, then constructing hyperedge diagram ++>。
In an embodiment, as shown in FIG. 2 (b), a hypergraph is constructedLater, by adding->The superside in (a) is regarded as a node, and a superside graph is constructed +.>To enhance hypergraph modeling capability, wherein +.>Is a hyperborder->Node set in->Is a hyperborder->Edge set in->To be +.>Strip superb->As a node->To be +.>Strip superb->As a node->Is a weight adjacency matrix defined based on the sequential relationship and the association relationship between the supersides.
(6-1) sequential relationship. In the intra-scale hypergraph, two hyperedges with time sequence nodes are connected to form a sequence relationship hyperedge graph, the sequence relationship hyperedge graph comprising node sets, edge sets and an adjacent matrix is constructed based on the sequence relationship, wherein the hyperedges of the intra-scale hypergraph are regarded as nodes of the sequence relationship hyperedge graph to form the node sets, the nodes of the sequence relationship hyperedge graph are constructed according to correlation to form the edge sets, the nodes of the sequence relationship hyperedge graph are used as the horizontal and vertical coordinates of the adjacent matrix, a value of 0 in the adjacent matrix indicates that the change of the node value connected with the former hyperedge in the intra-scale hypergraph does not affect the change of the node value connected with the latter hyperedge, and a value of 1 in the adjacent matrix indicates that the change of the node value connected with the former hyperedge in the intra-scale hypergraph affects the change of the node value connected with the latter hyperedge.
In an embodiment, in an intra-scale hypergraph, there is a sequential relationship connecting two hyperedges with time sequential nodes. As shown in fig. 4 (a), the superbConnect the first three nodes, < >>The next three nodes are connected. Because the time series is a set of data points arranged in chronological order, therefore +.>The change (e.g. increase, decrease or fluctuation) of the node value of the connection will be to +.>The node value of the connection has an effect.
Order theRepresenting a hyperedge graph constructed based on sequential relationships. Wherein the method comprises the steps ofRepresenting node set,/->For the number of intrascale supersides, +.>For the hyperedge set of intra-scale hypergraphs, +.>Representing a set of edges constructed based on sequential relationships. As shown in FIG. 4 (c), the adjacency matrix constructed based on the sequential relationship ++>The definition is as follows:
（15）。
(6-2) association relationship. In the inter-scale hypergraph and the mixed-scale hypergraph, two hyperedges with common nodes have an association relationship, an association relationship hypergraph comprising node sets, edge sets and an adjacent matrix is constructed based on the association relationship, wherein the hyperedges of the inter-scale hypergraph and the mixed-scale hypergraph are regarded as nodes of the association relationship hypergraph to form the node sets, the nodes of the association relationship hypergraph are constructed according to the correlation to form the edge sets, the nodes of the association relationship hypergraph are regarded as the horizontal coordinates and the vertical coordinates of the adjacent matrix, the value 0 in the adjacent matrix indicates that the change of the node values connected with the two hyperedges in the inter-scale hypergraph and the mixed-scale hypergraph does not have mutual influence, and the value 1 in the adjacent matrix indicates that the change of the node values connected with the two hyperedges in the inter-scale hypergraph and the mixed-scale hypergraph has mutual influence.
In the embodiment, in the inter-scale hypergraph and the mixed-scale hypergraph, an association relationship exists between two hyperedges with a common node. For example, sudden peaks in electricity usage may result in fluctuations in electricity usage on the current day as well as the current week. As shown in fig. 4 (b), the superbAnd->Because of the connection of the common node->But has an association relationship.
Order theAnd representing the hyperedge graph constructed based on the association relation. Wherein the method comprises the steps ofRepresenting node set,/->Is the number of inter-scale and mixed-scale superscales, +.>Is the hyperedge set of inter-scale hypergraph, +.>For a hyperedge set of a mixed-scale hypergraph,representing the edge set constructed based on the association relationship. As shown in (d) of FIG. 4, an adjacency matrix constructed based on the association relationship +.>The definition is as follows:
（16）。
specifically, filling and stacking an adjacency matrix constructed based on a sequential relation and an adjacency matrix constructed based on an association relation to obtain a total adjacency matrix, wherein the dimension of the total adjacency matrix is the sum of the nodes of the sequential relation superside graph and the nodes of the association relation superside graph.
In an embodiment, a total adjacency matrix constructed based on sequential relationships and associative relationshipsThe definition is as follows:
（17）
wherein,and->Is an all-zero matrix>Number of inter-scale and mixed-scale superscales +.>And the number of intrascale supersides +.>And the number of (3).
And 7, according to the constructed hypergraph and hyperedge graph, based on a three-stage message transmission mechanism, information of the node-hyperedge, the hyperedge-hyperedge and message transmission aggregation modes among the hyperedge-nodes is utilized, and interaction strength among time modes of different scales is learned.
The three-stage message delivery mechanism comprises the following specific steps:
(7-1) aggregating subsequencesNode embedding mapped to initialization ++>And obtaining an initialized superside embedding ++by aggregating the node embeddings connected by the superside>。
Giving a subsequence set obtained based on a hierarchical feature extraction moduleThe initialization node embedding +.>：
（18）
Wherein,for the mapping function, it may be implemented by a multi-layer perceptron (MLP),Afor the number of all nodes. After obtaining the initialized node embedding, the initialized superside embedding is obtained by aggregating the node information connected to the superside, as shown in fig. 2 (c). Specifically, for the->Strip superb->Its initialization node is embedded +.>Obtained by the following formula:
（19）
wherein,for passing beyond->Connected neighbor nodes.
(7-2) updating the superside embedding by means of the attention mechanism based on the constructed superside graph, and outputting the updated superside embedding。
Super-edge embedding after initializationAfter that, it is first mapped to the corresponding query +.>Key->Sum->Wherein->、/>And->Is a learnable mapping matrix. Wherein for->Is>Line vector->The updated superside is embedded with +.>The calculation can be made by the following formula:
（20）
wherein,is a key->Middle->Transpose of row vector, ">A value other than zero, i.e., one, +.>Is constant (I)>For the characteristic dimension of the sequence, +.>Is->Go->Concentration score function value of column->Is->Go->Concentration score function value of column->For attention score, ++>Is->Middle->Row vector->For the number of all superedges. In an embodiment, set->The value of (2) is 10000.
(7-3) dynamically updating the point-edge association matrix through the hypergraph attention mechanism based on the constructed hypergraph structureAnd updating node embedding through hypergraph convolution, and outputting updated node embedding ++>。
After the updated superedge embedding is obtained, the node embedding is updated by considering the relevant superedge connected with the corresponding node. First for each nodeAnd its related superedge, by node embedding +.>And updated superside embedding ++>Calculating an attention coefficient, and measuring interaction strength among time modes with different scales:
（21）
wherein,for the updated point-edge association matrix based on the attention mechanism +.>Embedding the node with the%>Representation of individual nodes->Is superb->Updated superside embedding ++>Is superb->Updated superside embedding ++>For the splicing operation of nodes and their related superedges, < >>Is a trainable multi-layer perceptron +.>For connecting to node->Can be selected by constructing hypergraph +.>Acquisition (I)>To activate the function.
Finally, the node embedding is updated by combining a multi-head attention mechanism (multi-head attention) and normalized hypergraph Laplace convolution, so that time mode information of different scales is aggregated:
（22）
wherein,for aggregating multiple head attention output, such as stitching or average pooling. />And->Respectively an updated point-edge association matrix and a learnable weight adjacency matrix +.>Is the number of the multi-head attention, +.>For the number of heads, ->To activate the function +.>Is the inverse square of the node degree matrix, +.>Is the inverse of the superside matrix, ++>Is the transpose of the point-side association matrix.
Step 8, embedding and splicing the last node of the subsequences with different scales to input a linear layer and output the futureStep electric load prediction value->。
Embedding the hierarchical supergraph updating nodes, embedding and splicing the last node of the sub-sequences with different scales, inputting the last node into a linear layer for fusion, and finally outputting the futureStep electric load prediction value->。
Step 9, calculating the prediction lossI.e. the true value corresponding to the training sample +.>And an actual modelOutput predictive value +.>Errors between them.
（23）
Wherein,and->Future +.>True and predicted values of steps.
Step 10, according to loss of all samples in the batchAnd adjusting network parameters, node embedding and over-edge embedding in the whole model.
Loss of all samples in the batch：
（24）
Wherein,is the%>Loss of individual samples, +.>The number of samples in each batch. In the embodiment of the invention, according to the loss +.>For a pair ofNetwork parameters, node embedding and scale embedding in the whole model (all parameters are noted as) And (3) adjusting and updating the following formula:
（25）/>
wherein,is the learning rate.
Step 11, repeating steps 3-10 until all batches of the training dataset are involved in model training.
Step 12, repeating the steps 3-11 until the specified iteration times are reached.
And 13, inputting the electric load sequence to be predicted into the trained model to obtain a prediction result.
In summary, the electric load prediction method based on the hierarchical hypergraph neural network provided by the embodiment of the invention models high-order interaction of time modes in a scale, among scales and in a mixed scale in electric load data through a hierarchical hypergraph structure, regards a superside as a node to construct the capability of the superside graph to enhance the high-order interaction between the modeling time modes of the model, learns the interaction strength between different scale time modes through a three-stage message transmission mechanism, further improves the prediction accuracy of an electric load prediction model, and has wide application prospects in the fields of electric power system operation, energy planning, energy efficiency management and the like.
The foregoing detailed description of the preferred embodiments and advantages of the invention will be appreciated that the foregoing description is merely illustrative of the presently preferred embodiments of the invention, and that no changes, additions, substitutions and equivalents of those embodiments are intended to be included within the scope of the invention.
Claims (9)
1. The electric load prediction method based on the hierarchical hypergraph neural network is characterized by comprising the following steps of:
s1, preprocessing electric load data and constructing a training sample;
s2, mapping the sequence of the training sample into subsequences with different time scales by using a layering characteristic extraction module;
s3, in a hypergraph-hyperedge graph construction module, electric load data at different moments in a subsequence are regarded as nodes, and a hierarchical hypergraph is constructed according to a time proximity rule so as to model high-order interactions among intra-scale, inter-scale and mixed-scale time modes; constructing a hypergraph based on the sequence relation and the association relation among the nodes by taking the hyperedge of the hypergraph as the node;
the construction of the hypergraph based on the sequence relation and the association relation between the nodes by regarding the hyperedge of the hypergraph as the node comprises the following steps:
in the intra-scale hypergraph, two hyperedges with time sequence nodes are connected to form a sequence relation hyperedge graph comprising node sets, edge sets and an adjacent matrix based on the sequence relation, wherein the hyperedges of the intra-scale hypergraph are regarded as nodes of the sequence relation hyperedge graph to form the node sets, the nodes of the sequence relation hyperedge graph form the edge sets according to the correlation, the nodes of the sequence relation hyperedge graph are regarded as the horizontal and vertical coordinates of the adjacent matrix, the value of 0 in the adjacent matrix indicates that the two hyperedges in the intra-scale hypergraph have no sequence relation, and the value of 1 in the adjacent matrix indicates that the two hyperedges in the intra-scale hypergraph have the sequence relation;
in the inter-scale hypergraph or the mixed-scale hypergraph, two hyperedges with common nodes have an association relationship, an association relationship hypergraph comprising node sets, edge sets and an adjacent matrix is constructed based on the association relationship, wherein the hyperedges of the inter-scale hypergraph or the mixed-scale hypergraph are regarded as nodes of the association relationship hypergraph to form the node sets, the nodes of the association relationship hypergraph form the edge sets according to the correlation construction edges, the nodes of the association relationship hypergraph are regarded as horizontal coordinates and vertical coordinates of the adjacent matrix, a value of 0 in the adjacent matrix indicates that the association relationship exists between the two hyperedges in the inter-scale hypergraph or the mixed-scale hypergraph, and a value of 1 in the adjacent matrix indicates that the association relationship does not exist between the two hyperedges in the inter-scale hypergraph or the mixed-scale hypergraph;
filling and stacking the adjacency matrix constructed based on the sequential relation and the adjacency matrix constructed based on the association relation to obtain a total adjacency matrix, wherein the dimension of the total adjacency matrix is the sum of the nodes of the sequential relation superside graph and the nodes of the association relation superside graph;
s4, in the three-stage message transfer module, the time mode information of different scales is aggregated through a node-superside, a superside-superside and a three-stage message transfer mechanism among the superside-nodes, and the interaction strength among the time modes of different scales is learned;
s5, fusing time mode information of different scales, carrying out electric load prediction, and training an electric load prediction model based on all training samples, wherein the electric load prediction model comprises a layering characteristic extraction module, a hypergraph-hyperedge graph construction module and a three-stage message transmission module;
s6, inputting the electric load sequence to be predicted into a trained electric load prediction model to obtain a prediction result.
2. The hierarchical hypergraph neural network-based power load prediction method according to claim 1, wherein in S1, the preprocessing the power load data and constructing a training sample includes:
preprocessing given electric load data, including outlier processing and normalization processing; dividing the preprocessed power load data through a sliding time window to obtain a training data set; the training data set is batched according to the fixed batch size, and training samples are selected according to batches.
3. The method for predicting the electrical load based on the hierarchical hypergraph neural network according to claim 1, wherein in S2, the mapping the sequence of training samples into subsequences with different time scales by using the hierarchical feature extraction module comprises:
the sequence of the training sample is input into a layering characteristic extraction module, aggregation functions are utilized to generate subsequences with different time scales by setting aggregation windows with different sizes, the length of the next time scale subsequence is the length of the last time scale subsequence divided by the size of the aggregation window corresponding to the last time scale, and finally all the subsequences form a subsequence set.
4. The method for predicting electrical loads based on hierarchical hypergraph neural networks according to claim 1, wherein in S3, the modeling of the hierarchical hypergraph to model high-order interactions within-scale, inter-scale, and between-mixed-scale time patterns according to the temporal proximity rule by regarding the electrical load data at different times in the sub-sequence as nodes comprises:
taking the power load data at different moments in the subsequence as nodes, constructing an original intra-scale hypergraph, an original inter-scale hypergraph and an original mixed-scale hypergraph according to a time proximity rule, and respectively constructing hypergraphs of each typekJump connection to aggregate information in different neighbor rangeskJump-scale hypergraph,kInter-scale hopping hypergraphkJump hybrid scale hypergraph, whereinkFor the time span between two adjacent nodes, finally obtaining the graph containing the original intra-scale hypergraph andkan intra-scale hypergraph that jumps to an intra-scale hypergraph, comprising an original inter-scale hypergraph andkinter-scale hypergraphs that skip inter-scale hypergraphs, comprising an original mixed-scale hypergraph andka hybrid scale hypergraph that hops the hybrid scale hypergraph.
5. The method for predicting electrical loads based on hierarchical hypergraph neural network as set forth in claim 4, wherein the intra-scale hypergraph is obtained by constructing original hypergraphs by constructing original connections between adjacent nodes in the same scale according to correlation to obtain a plurality of hyperedges, and constructing nodes with time spans and correlation information in the same scalekJump connection to obtain multiple superedge constructionkJumping the hypergraph in the scale, and summing the hypergraph in the original scalekAnd obtaining the intra-scale hypergraph by splicing the intra-scale hypergraph.
6. The method for predicting electrical loads based on hierarchical hypergraph neural network according to claim 4, wherein the inter-scale hypergraph is obtained by constructing original connections between adjacent nodes and nodes of a next scale in the same scale according to correlation to obtain a plurality of hyperedges, and constructing an original inter-scale hypergraph between nodes having time spans and correlation information in the same scale and nodes of the next scalekJump connection to obtain multiple superedge constructionkJumping the hypergraph between scales, summing the hypergraph between original scaleskAnd obtaining the inter-scale hypergraph by splicing the inter-scale hypergraph.
7. The method for predicting electrical loads based on hierarchical hypergraph neural network as set forth in claim 4, wherein the hybrid-scale hypergraph is constructed by constructing a plurality of hyperedges for establishing original connections between nodes of each scale according to correlations, wherein nodes within the same scale to which the hyperedges are connected are adjacent nodes, and establishing between nodes of each scalekJump connection to obtain multiple superedge constructionkJumping mixed scale hypergraph, wherein nodes in the same scale connected by hyperedges are nodes with time span and correlation information, and combining the original mixed scale hypergraph withkAnd obtaining the mixed scale hypergraph by splicing the mixed scale hypergraph.
8. The method for predicting the electrical load based on the hierarchical hypergraph neural network according to claim 1, wherein in S4, in the three-phase message passing module, the interaction strength between the time modes of different scales is learned by aggregating the time mode information of different scales through a node-hyperedge, a hyperedge-hyperedge and a three-phase message passing mechanism between the hyperedge and the node, including:
s41, mapping the subsequence into an initialized node embedding, and obtaining the initialized superside embedding by aggregating the node embedding connected with the superside;
s42, updating the superside embedding through an attention mechanism based on the constructed superside graph, and outputting the updated superside embedding;
s43, dynamically updating the point-edge association matrix through a hypergraph attention mechanism based on the constructed hierarchical hypergraph, updating node embedding through hypergraph convolution, and outputting updated node embedding.
9. The method for predicting the electrical load based on the hierarchical hypergraph neural network according to claim 8, wherein in S6, the time mode information of different scales is fused to perform electrical load prediction, after the updated node is embedded, the last node of the sub-sequences of different scales is embedded and spliced to be input into a linear layer to be fused, and finally, the electrical load predicted value is output.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311123122.5A CN116845889B (en) | 2023-09-01 | 2023-09-01 | Hierarchical hypergraph neural network-based power load prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311123122.5A CN116845889B (en) | 2023-09-01 | 2023-09-01 | Hierarchical hypergraph neural network-based power load prediction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116845889A CN116845889A (en) | 2023-10-03 |
| CN116845889B true CN116845889B (en) | 2023-12-22 |
Family
ID=88172893
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311123122.5A Active CN116845889B (en) | 2023-09-01 | 2023-09-01 | Hierarchical hypergraph neural network-based power load prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116845889B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117290800B (en) * | 2023-11-24 | 2024-01-26 | 华东交通大学 | Timing sequence anomaly detection method and system based on hypergraph attention network |
| CN117688975B (en) * | 2024-02-02 | 2024-05-14 | 南京信息工程大学 | Meteorological event prediction method and system based on evolution rule mining |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112417219A (en) * | 2020-11-16 | 2021-02-26 | 吉林大学 | Hyper-graph convolution-based hyper-edge link prediction method |
| CN113988415A (en) * | 2021-10-28 | 2022-01-28 | 河北工业大学 | Medium-and-long-term power load prediction method |
| CN114169394A (en) * | 2021-11-04 | 2022-03-11 | 浙江大学 | Multi-variable time series prediction method for multi-scale adaptive graph learning |
| CN114463596A (en) * | 2021-12-28 | 2022-05-10 | 苏州科技大学 | Small sample image identification method, device and equipment of hypergraph neural network |
| CN114547325A (en) * | 2022-01-14 | 2022-05-27 | 北京帝测科技股份有限公司 | Probabilistic hypergraph-driven geoscience knowledge graph reasoning optimization system and method |
| CN115049629A (en) * | 2022-06-27 | 2022-09-13 | 太原理工大学 | Multi-mode brain hypergraph attention network classification method based on line graph expansion |
| CN115423076A (en) * | 2022-08-17 | 2022-12-02 | 中国人民解放军国防科技大学 | Directed hypergraph chain prediction method based on two-step framework |
| CN115983341A (en) * | 2022-12-27 | 2023-04-18 | 燕山大学 | Node classification method based on relation aggregation hypergraph |
| WO2023098098A1 (en) * | 2021-12-02 | 2023-06-08 | 南京邮电大学 | Tag-aware recommendation method based on attention mechanism and hypergraph convolution |
| CN116340646A (en) * | 2023-01-18 | 2023-06-27 | 云南师范大学 | Recommendation method for optimizing multi-element user representation based on hypergraph motif |
| CN116363874A (en) * | 2023-03-20 | 2023-06-30 | 南京理工大学 | Urban traffic hypergraph convolution prediction method integrating multimode high-order semantic correlation |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9208278B2 (en) * | 2013-06-26 | 2015-12-08 | Synopsys, Inc. | Clustering using N-dimensional placement |
| US20220414520A1 (en) * | 2021-06-23 | 2022-12-29 | Servicenow Canda Inc. | Method and system for training model to perform link prediction in knowledge hypergraph |
-
2023
- 2023-09-01 CN CN202311123122.5A patent/CN116845889B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112417219A (en) * | 2020-11-16 | 2021-02-26 | 吉林大学 | Hyper-graph convolution-based hyper-edge link prediction method |
| CN113988415A (en) * | 2021-10-28 | 2022-01-28 | 河北工业大学 | Medium-and-long-term power load prediction method |
| CN114169394A (en) * | 2021-11-04 | 2022-03-11 | 浙江大学 | Multi-variable time series prediction method for multi-scale adaptive graph learning |
| WO2023098098A1 (en) * | 2021-12-02 | 2023-06-08 | 南京邮电大学 | Tag-aware recommendation method based on attention mechanism and hypergraph convolution |
| CN114463596A (en) * | 2021-12-28 | 2022-05-10 | 苏州科技大学 | Small sample image identification method, device and equipment of hypergraph neural network |
| CN114547325A (en) * | 2022-01-14 | 2022-05-27 | 北京帝测科技股份有限公司 | Probabilistic hypergraph-driven geoscience knowledge graph reasoning optimization system and method |
| CN115049629A (en) * | 2022-06-27 | 2022-09-13 | 太原理工大学 | Multi-mode brain hypergraph attention network classification method based on line graph expansion |
| CN115423076A (en) * | 2022-08-17 | 2022-12-02 | 中国人民解放军国防科技大学 | Directed hypergraph chain prediction method based on two-step framework |
| CN115983341A (en) * | 2022-12-27 | 2023-04-18 | 燕山大学 | Node classification method based on relation aggregation hypergraph |
| CN116340646A (en) * | 2023-01-18 | 2023-06-27 | 云南师范大学 | Recommendation method for optimizing multi-element user representation based on hypergraph motif |
| CN116363874A (en) * | 2023-03-20 | 2023-06-30 | 南京理工大学 | Urban traffic hypergraph convolution prediction method integrating multimode high-order semantic correlation |
Non-Patent Citations (3)
| Title |
|---|
| Geographic-Semantic-Temporal Hypergraph Convolutional Network for Traffic Flow Prediction;Kesu Wang et al.;《 2020 25th International Conference on Pattern Recognition (ICPR)》;第5444-5450页 * |
| 基于超图模型的大规模门级网表层次化聚类算法;蒿杰 等;《计算机辅助设计与图形学学报》;第21卷(第01期);第44-52页 * |
| 多通道图神经网络的层次化融合模型用于增强会话的推荐;岳彩梦 等;《小型微型计算机系统》;第1-11页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116845889A (en) | 2023-10-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Hao et al. | A novel two-stage forecasting model based on error factor and ensemble method for multi-step wind power forecasting | |
| CN116845889B (en) | Hierarchical hypergraph neural network-based power load prediction method | |
| Tian et al. | Short-term wind speed prediction based on improved PSO algorithm optimized EM-ELM | |
| CN111489036B (en) | Resident load prediction method and device based on electrical load characteristics and deep learning | |
| CN111680841B (en) | Short-term load prediction method, system and terminal equipment based on principal component analysis | |
| Liu et al. | Heating load forecasting for combined heat and power plants via strand-based LSTM | |
| CN113554466A (en) | Short-term power consumption prediction model construction method, prediction method and device | |
| CN111882157A (en) | Demand prediction method and system based on deep space-time neural network and computer readable storage medium | |
| Sha et al. | Development of a multi-granularity energy forecasting toolkit for demand response baseline calculation | |
| Chen et al. | MultiCycleNet: multiple cycles self-boosted neural network for short-term electric household load forecasting | |
| CN113591368A (en) | Comprehensive energy system multi-energy load prediction method and system | |
| Wang et al. | A novel wind power prediction model improved with feature enhancement and autoregressive error compensation | |
| CN111680939A (en) | Enterprise re-work and re-production degree monitoring method based on artificial intelligence | |
| Zhao et al. | Application of a novel hybrid accumulation grey model to forecast total energy consumption of Southwest Provinces in China | |
| Zhang et al. | Ensemble learning‐based modeling and short‐term forecasting algorithm for time series with small sample | |
| Liang et al. | A wind speed combination forecasting method based on multifaceted feature fusion and transfer learning for centralized control center | |
| CN114611757A (en) | Electric power system short-term load prediction method based on genetic algorithm and improved depth residual error network | |
| Tang et al. | Short-term power load forecasting based on VMD-Pyraformer-Adan | |
| CN117390550A (en) | Low-carbon park carbon emission dynamic prediction method and system considering emission training set | |
| Ajagunsegun et al. | Machine Learning-Based System for Managing Energy Efficiency of Public Buildings: An Approach towards Smart Cities | |
| Cao et al. | A hybrid electricity load prediction system based on weighted fuzzy time series and multi-objective differential evolution | |
| Zou et al. | Predicting the electric power consumption of office buildings based on dynamic and static hybrid data analysis | |
| Zhang et al. | A novel transfer deep learning model with reinforcement-learning-based hyperparameter optimization for short-term load forecasting during the COVID-19 pandemic | |
| Prashanthi et al. | A comparative study of the performance of machine learning based load forecasting methods | |
| CN114861967A (en) | Power load prediction method, system, device and storage medium |
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 |