CN116151459A - Power grid flood prevention risk probability prediction method and system based on improved Transformer - Google Patents
Power grid flood prevention risk probability prediction method and system based on improved Transformer Download PDFInfo
- Publication number
- CN116151459A CN116151459A CN202310178028.3A CN202310178028A CN116151459A CN 116151459 A CN116151459 A CN 116151459A CN 202310178028 A CN202310178028 A CN 202310178028A CN 116151459 A CN116151459 A CN 116151459A
- Authority
- CN
- China
- Prior art keywords
- flood prevention
- feature
- data
- power grid
- input
- 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
Links
- 230000002265 prevention Effects 0.000 title claims abstract description 82
- 238000000034 method Methods 0.000 title claims abstract description 41
- 230000007246 mechanism Effects 0.000 claims abstract description 27
- 230000006872 improvement Effects 0.000 claims abstract description 21
- 238000004458 analytical method Methods 0.000 claims abstract description 18
- 230000004927 fusion Effects 0.000 claims abstract description 14
- 230000004048 modification Effects 0.000 claims abstract description 6
- 238000012986 modification Methods 0.000 claims abstract description 6
- 239000011159 matrix material Substances 0.000 claims description 40
- 239000013598 vector Substances 0.000 claims description 36
- 238000004364 calculation method Methods 0.000 claims description 20
- 230000003068 static effect Effects 0.000 claims description 17
- 238000012544 monitoring process Methods 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 10
- 238000010606 normalization Methods 0.000 claims description 7
- 238000007781 pre-processing Methods 0.000 claims description 7
- 230000009466 transformation Effects 0.000 claims description 7
- 238000004140 cleaning Methods 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 5
- 230000002159 abnormal effect Effects 0.000 claims description 4
- 230000036962 time dependent Effects 0.000 claims description 4
- 230000002708 enhancing effect Effects 0.000 abstract description 6
- 238000010586 diagram Methods 0.000 description 6
- 238000012545 processing Methods 0.000 description 4
- 238000013136 deep learning model Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000007405 data analysis Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000004321 preservation Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 230000001808 coupling effect Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000013058 risk prediction model Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000012876 topography Methods 0.000 description 1
- 230000007723 transport mechanism Effects 0.000 description 1
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
- 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
- G06Q10/0635—Risk analysis of enterprise or organisation activities
-
- 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
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
- Y02A10/40—Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Development Economics (AREA)
- Health & Medical Sciences (AREA)
- Educational Administration (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a power grid flood prevention risk prediction method and a power grid flood prevention risk prediction system based on an improved converter, wherein the prediction method comprises the following steps: acquiring flood prevention data; carrying out power grid flood prevention prediction early warning influence factor identification analysis on flood prevention data; modifying the Transformer network framework using a feature enhancement modification strategy based on a gating selection mechanism and embedded coding; then, an explicit sparse attention improvement strategy is used for improving a Transformer network framework; and carrying out power grid flood prevention risk prediction by using the improved converter network framework. According to the invention, the network framework based on the Transformer is used for enhancing the overall network performance and improving the accuracy of flood prevention risk probability prediction by optimizing the feature selection module, the feature fusion module and the attention module.
Description
Technical Field
The invention belongs to the field of data analysis in the power industry, and particularly relates to a power grid flood prevention risk probability prediction method and system based on an improved Transformer.
Background
The accurate prediction of the flood prevention risk probability of the transformer substation is of great significance to the improvement of the flood prevention capability of the power grid and the healthy flood prevention early warning system. The flood prevention influencing factors of the transformer substation not only have dynamic meteorological data, but also have static data such as the volume of a water collecting well of the transformer substation, the drainage amount of a water pump, the storage of flood prevention materials, the topography, the hydrologic characteristics and the like, and the dynamic and static combination brings difficulty to the prediction of the traditional flood prevention risk.
Conventional time series prediction methods represented by the autoregressive integrated moving average (ARIMA) model have strong interpretations, but they are mainly applicable to univariate prediction problems, which limit their application to complex multivariate time series data. Different from the independent prediction of single or small amount of time sequences, the machine learning-based method fully utilizes large-scale time sequence data in the existing application scene through feature processing, a more accurate model can be fitted, but the machine learning method has limited capability in the aspect of processing original data, professional knowledge in related fields is required to convert the original data into related representations or features, the quality of the original data representation directly influences the performance of the machine learning model, compared with the data-driven deep learning model which can learn the related representations and features of the data without a large amount of manual feature engineering, the method is suitable for the substation flood prevention risk prediction scene under the large-scale data by combining simple nonlinear function modules, converting low-level input into high-level abstract feature representation and automatically capturing complex relations in the original high-dimensional data. However, a large amount of data is required to train the deep learning model as a training basis, the solvability of the deep learning model is not strong, the result obtained by training the data of a single station is difficult to apply to the flood prevention risk probability prediction of other substations, and the generalization capability is not strong.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a power grid flood prevention risk probability prediction method and system based on an improved transducer, which solve the problems that flood prevention data is complex, has a certain difference with text data, is difficult to directly use on a transducer model, the transducer model cannot fully utilize time stamp information of a time sequence, and the complexity of a self-attention module in the transducer model in calculation time and memory consumption is high in the prior art.
In order to achieve the above purpose, the invention adopts the following technical scheme:
an improved transducer-based power grid flood prevention risk prediction method comprises the following steps:
acquiring flood prevention data;
carrying out power grid flood prevention prediction early warning influence factor identification analysis on flood prevention data;
modifying the Transformer network framework using a feature enhancement modification strategy based on a gating selection mechanism and embedded coding;
then, an explicit sparse attention improvement strategy is used for improving a Transformer network framework;
and carrying out power grid flood prevention risk prediction by using the improved converter network framework.
Preferably, before the identifying and analyzing the flood prevention data according to the power grid flood prevention prediction early warning influence factors, the method further includes:
after flood control data are acquired, data preprocessing is carried out, and then identification analysis of power grid flood control prediction early warning influence factors is carried out;
the data preprocessing comprises data cleaning and data normalization;
the data cleaning is used for removing abnormal values in the data set and filling missing values;
the data normalization is used to make the data features of the same metric.
Preferably, the identifying and analyzing the flood prevention data for the power grid flood prevention prediction and early warning influence factors includes:
dividing monitoring information of a transformer substation into meteorological monitoring information, geographical environment information and site self information;
the geographic environment information and the site self information are static covariate characteristics of the transformer substation;
the weather monitoring information is a dynamic covariate characteristic.
Preferably, the feature enhancement improvement strategy using a gating selection mechanism and embedded coding comprises: the main body input feature selection strategy based on the gating mechanism and the time stamp feature fusion strategy based on the embedded coding.
Preferably, the body input feature selection policy based on the gating mechanism includes:
taking the core meteorological features as main body input feature vectors, taking the static covariate features of the site as optional context feature vectors, and taking the main body input feature vectors and the optional context feature vectors as input features;
the input characteristics are operated through a gating residual error network module and softmax to obtain the selection weight of the input variable;
performing weighted assignment on the selection weight of the input variable and the input feature to obtain example feature selection of the input feature;
a feature variable selection network is constructed for the static covariates and time-dependent sequence variables applied to the instance feature selection.
Preferably, the embedding encoding-based time stamp feature fusion strategy comprises:
s1, taking an input sequence with a reserved t moment as an input transformation sequence vector;
s2, embedding a local position feature vector of a storage sequence by using a fixed position;
s3, representing the global timestamp feature vector by using a learnable time code;
s4, mapping the input transformation sequence, the local position feature and the global time stamp feature vector to the same feature dimension by using a one-dimensional convolution filter with the kernel size of 3 and the step distance of 1, and fusing to obtain the enhanced input feature embedded code.
Preferably, the using an explicit sparse attention improvement strategy is implemented by a selection strategy employing top-k, comprising:
obtaining an original attention score matrix by using keys in a self-attention mechanism and query calculation;
multiplying the original attention score matrix with a threshold matrix, and then calculating by a sign function to obtain an original position mask matrix;
replacing 1 in the original position mask matrix with 0 and replacing 0 with minus infinity to obtain a position mask matrix;
and adding the original attention score matrix and the position mask matrix to obtain an attention score matrix.
An improved fransformer-based power grid flood prevention risk prediction system, comprising: the power grid flood prevention risk prediction method based on the improved Transformer is realized when the processor executes the computer program.
A computer-readable storage medium storing computer-executable instructions for causing a computer to perform the improved fransformer-based grid flood prevention risk probability prediction method.
Although the transducer has excellent performance as a representative of deep learning in recent years, the transducer model is designed for problems related to natural language processing, and although both text sequences and time sequences are sequence data, there are characteristic differences, such as timestamp information. From the above principle analysis, it is known that the transducer model performs position coding using a sine and cosine function, then adds it to the input sequence embedding and feeds it to the prediction model, and although this approach can extract some position information from the time sequence, it cannot make full use of the time stamp information of the time sequence, so that it is necessary to code the time stamp as an additional position code. Meanwhile, in order to fully utilize the static basic characteristics of the transformer substation, the implicit action relation between the meteorological factors and the static characteristics is learned, and the static characteristics are required to be interactively selected. In addition, in consideration of complexity of the self-attention module in the transducer model in terms of calculation time and memory consumption, in order to reduce requirements on hardware equipment in power grid early warning management and improve model efficiency, improvement on the self-attention model is required. In summary, the invention aims at the defects in the transducer model, combines the characteristics of the prediction object to carry out algorithm adaptability improvement, extracts high-dimensional coupling sequence characteristics on the premise of comprehensively considering the coupling effect of the multidimensional influence factors of flood disasters, constructs a flood risk prediction model, realizes the prediction and early warning of the flood risk of the Transformer substation, and is detailed as follows:
the invention provides a power grid flood prevention risk prediction method based on an improved transducer, which uses an improved transducer network frame to predict the power grid flood prevention risk probability, and comprises the following steps: a optimization model structure based on a Transformer network framework is selected according to flood prevention data characteristics, and a characteristic enhancement improvement strategy and an explicit sparse attention improvement strategy based on a gating selection mechanism and embedded codes are used. According to the method, a power grid flood prevention prediction early warning influence factor identification analysis strategy is provided, and the characteristic factors of power grid flood prevention data are utilized efficiently; providing a characteristic enhancement strategy based on a gating selection mechanism and embedded codes, enhancing information extraction of data characteristics, and enhancing input of a prediction model; the explicit sparse attention improvement strategy is provided, irrelevant information is effectively removed, complexity of the self-attention module in calculation time and memory consumption is reduced, and good performance is shown in the aspect of power grid flood prevention prediction.
The invention provides an improved Transformer-based power grid flood prevention risk prediction method, which is a network framework based on Transformer, and is characterized in that a feature selection module, a feature fusion module and an attention module are optimized, so that the overall network performance is enhanced, the accuracy of flood prevention risk probability prediction is improved, and the problems that flood prevention data is complex, has a certain difference with text data, is difficult to directly use on a Transformer model, time stamp information of a time sequence cannot be fully utilized by the Transformer model, and the complexity of a self-attention module in the Transformer model in calculation time and memory consumption is high in the prior art are solved.
Drawings
FIG. 1 is a flow chart diagram of an improved transducer-based power grid flood prevention risk probability prediction method;
FIG. 2 is a block flow diagram of a subject input feature selection strategy based on gating mechanisms of the present invention;
FIG. 3 is a flow chart of the embedded code based timestamp feature fusion strategy of the present invention;
FIG. 4 is a block flow diagram of the present invention using an explicit sparse attention improvement strategy.
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 specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
It should be noted that although functional block diagrams are depicted as block diagrams, and logical sequences are shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the block diagrams in the system. The terms first, second and the like in the description and in the claims and in the above-described figures, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order.
The invention provides a power grid flood prevention risk prediction method based on an improved Transformer. According to the method, a power grid flood prevention prediction early warning influence factor identification analysis strategy is provided, and the characteristic factors of power grid flood prevention data are utilized efficiently; providing a characteristic enhancement strategy based on a gating selection mechanism and embedded codes, enhancing information extraction of data characteristics, and enhancing input of a prediction model; the explicit sparse attention improvement strategy is provided, irrelevant information is effectively removed, complexity of the self-attention module in calculation time and memory consumption is reduced, and good performance is shown in the aspect of power grid flood prevention prediction.
Embodiments of the present invention will be further described with reference to the accompanying drawings.
The overall architecture of the transducer is divided into four modules: the device comprises an input module, an encoding module, a decoding module and an output module. The input module comprises: the source data embedded layer and the position encoder thereof adopt a data input coding strategy; the coder module and the decoder module comprise: the system comprises a self-attention module, a normalization layer, a feedforward full-connection sub-layer and a residual error connection module; the output module includes: linear layer and softmax layer.
Referring to fig. 1, fig. 1 is a flowchart of an improved transporter based power grid flood prevention risk prediction method according to an embodiment of the present invention, where the improved transporter based power grid flood prevention risk prediction method includes, but is not limited to, steps S110 to S150.
Step S110, flood control data are obtained;
step S120, carrying out identification analysis on the flood control prediction early warning influence factors of the power grid on the flood control data;
step S130, improving the flood control data input coding strategy by using a characteristic enhancement improvement strategy based on a gating selection mechanism and embedded coding;
step S140, the self-attention module is improved by using the explicit sparse attention improvement strategy;
and step S150, predicting the risk probability of flood control of the power grid by using the improved converter network framework obtained in the step S140.
Further, before the power grid flood prevention prediction early warning influence factor identification analysis is performed on the flood prevention data, the method further comprises the following steps:
after flood control data are acquired, data preprocessing is carried out, and then identification analysis of power grid flood control prediction early warning influence factors is carried out;
and carrying out data preprocessing on the flood prevention data, including data cleaning and data normalization.
The data cleaning is used for removing abnormal values in the data set and filling missing values; the data normalization is used to make the data features of the same metric.
Further, carrying out identification analysis on the power grid flood prevention prediction early warning influence factors on the flood prevention data, wherein the identification analysis comprises the following steps:
the monitoring information of the transformer substation is divided into three aspects, namely weather monitoring information, geographical environment information and site self information, through identification and analysis of the flood prevention influence factors of the transformer substation; the geographic environment and the site self information are static covariate characteristics of the transformer substation; meteorological monitoring information is categorized into dynamic covariate characteristics.
Further, feature enhancement refinement strategies based on gating selection mechanisms and embedded coding are used, including: the main body input feature selection strategy based on the gating mechanism and the time stamp feature fusion strategy based on the embedded coding.
Still further, referring to fig. 2, the gating mechanism-based principal input feature selection strategy includes the following steps, but is not limited to steps S210 to S240:
step S210, taking the core meteorological features as main input feature vectors, taking the static covariate features of the site as optional context feature vectors, and taking the main input feature vectors and the optional context feature vectors as input features;
step S220, the input characteristics are operated through a gate residual network module and softmax to obtain the selection weight of the input variable;
step S230, carrying out weighted assignment on the selection weights of the input variables and the input features to obtain example feature selection of the input features;
step S240, constructing a feature variable selection network of static covariates and time-dependent sequence variables applied to instance feature selection.
Still further, referring to fig. 3, the embedded encoding-based timestamp feature fusion strategy includes the following steps, but is not limited to steps S310 to S340:
step S310, taking the input sequence with the t moment reserved as an input transformation sequence vector;
step S320, embedding the local position feature vector of the preservation sequence by using the fixed position;
step S330, the global timestamp feature vector is characterized by using a learnable time code;
and step S340, mapping the input transformation sequence, the local position feature and the global time stamp feature vector to the same feature dimension by using a one-dimensional convolution filter with a kernel size of 3 and a step pitch of 1, and fusing to obtain the enhanced input feature embedded code.
Further, referring to fig. 4, the use of an explicit sparse attention improvement strategy is achieved by employing a top-k selection strategy comprising the steps of:
step S410, obtaining an original attention score matrix by calculation by using keys in a self-attention mechanism and a query;
step S420, multiplying the original attention score matrix with a threshold matrix, and then calculating by a sign function to obtain an original position mask matrix;
step S430, replacing 1 in the original position mask matrix with 0, and replacing 0 with minus infinity to obtain a position mask matrix;
step S440, adding the original attention score matrix to the position mask matrix to obtain an attention score matrix.
In one embodiment, first, flood control data is obtained; then, carrying out identification analysis on the power grid flood prevention prediction early warning influence factors on the data; then, a feature enhancement improvement strategy based on a gating selection mechanism and embedded codes is used; explicit sparse attention improvement strategies are then used; finally, the improved transducer network frame is used for predicting the flood prevention risk probability of the power grid, so that the high-efficiency and accurate prediction of the flood prevention risk probability of the power grid is realized, and the method is detailed as follows:
step S1: data analysis
Acquiring flood prevention data, performing data preprocessing on the data, removing abnormal values in a data set, filling in missing values, and enabling data features to have the same measurement scale; and then, carrying out identification analysis on the power grid flood prevention prediction early warning influence factors on the data, dividing the monitoring information of the transformer substation into three aspects, namely dynamic covariate weather monitoring information serving as core weather data, geographical environment information serving as static covariate and site self information, and further selecting related variables to improve model prediction accuracy.
Step S2: feature selection
The method comprises the steps of using a main body input characteristic selection strategy based on a gating mechanism and a timestamp characteristic fusion strategy based on embedded coding to enhance the input of a prediction model, wherein the main body input characteristic selection strategy based on the gating mechanism is as follows:
step S21: taking the core meteorological features as main input feature vectors a, and taking the static covariate features of the site as optional context feature vectors c;
step S22: the main body input feature vector and the optional context feature vector are subjected to a gating residual error network module and softmax operation to obtain the selection weight w of the input variable xt Selecting a weight w xt Is realized as follows:
w xt =softmax(GRN(ξ t ))
in xi t The characteristic vector is input at the moment t; the calculation of the gate-control residual network GRN is implemented as:
GRN(a,c)=LayerNorm(a+GLU(η 1 ))
η 1 =W 1 η 2 +b 1
η 2 =ELU(W 2 a+W 3 c+b 2 )
wherein a is a main body input feature vector; c is an optional context feature vector; ELU is an activation function; w (W) i Weight parameters which can be learned for the model; b i As a bias parameter, eta i Is an intermediate variable.
The gating linear unit GLU is a specific implementation of a gating layer, the input of a model is adjusted through a gating mechanism, the flexibility of the model is improved, and the calculation mode of the GLU is as follows:
GLU(γ)=σ(W 4 γ+b 4 )⊙(W 5 γ+b 5 )
in the formula, GLU is a gating linear unit, gamma is the input of a gating layer, and sigma is a Sigmoid activation function;
step S23: the selection weight of the input variable is subjected to weighted assignment with the input feature to obtain the example feature selection of the input feature at the moment t Is realized as follows:
step S24: a feature variable selection network is constructed for the static covariates and time-dependent sequence variables applied to the instance feature selection.
Step S3: feature fusion
The method comprises the steps of using a main body input feature selection strategy based on a gating mechanism and a time stamp feature fusion strategy based on embedded coding to enhance the input of a prediction model, wherein the time stamp feature fusion strategy based on the embedded coding is as follows:
step S31: the input sequence with t moment is reserved as the characteristic vector xi input by the input transformation sequence for t moment t ;
Step S32: local position feature PE using fixed position embedding preservation sequences pos The calculation formula of the position coding function is as follows:
wherein pos represents the position of the original data in the sequence, and d represents the dimension of the multidimensional feature after being subjected to the mapping of the input vector embedding layer;
step S33: the characteristic of the global time stamp is characterized by using a learnable time code, and the specific coding parameter calculation mode is as follows:
in TE i Representing a class i timestamp encoding; t is t i And T i Representing the time point and period under the time stamp of the type, taking a month as an example, T represents the sampling time month of the data to be encoded, and T is period 12;
step S34: for t timeExample feature selection of an incoming featureLocal position feature vector PE pos And a global timestamp feature vector TE pos Using one-dimensional convolution filter with kernel size of 3 and stride of 1, the same feature dimension d model And fusing to obtain the embedded code χ of the enhanced input feature t The specific calculation of the module is as follows:
where α is a factor that balances the size between the projection of the input sequence and the local and global embeddings, if the input sequence is normalized, it is recommended that α have a value of 1.
Step S4: self-attention thinning
Step S41: the original attention score matrix P is calculated by using keys in the self-attention mechanism and the query, and the calculation of the original attention score matrix P is implemented as follows:
wherein Q and K are queries and keys in the attention mechanism; d, d k Is the characteristic dimension of each attention head;
step S42: multiplying the original attention score matrix P with a threshold matrix T, and then calculating by a sign function to obtain an original position mask matrix M, wherein the calculation of the original position mask matrix M is realized as follows:
M=sign(PT);
step S43: replacing 1 in the original position mask matrix M with 0, and replacing 0 with minus infinity to obtain a position mask matrix M';
step S44: adding the original attention score matrix and the position mask matrix to obtain a first output Y, and calculating the first output Y through a softmax function to obtain an attention score matrix A, wherein the calculation of the attention score matrix A is realized as follows:
A=softmax(PM′)
step S45: the standardized attention score matrix is obtained through the above process, and the final self-attention output C is obtained by calculation by combining the value vector V obtained by the source context mapping, wherein the calculation formula is as follows
C=AV。
Step S5: network output
And selecting a Transformer as a basic network frame, adopting an optimization and improvement strategy from step S2 to step S4 to optimize the feature selection module, the feature fusion module and the attention module, enhancing the overall network performance and reducing the calculation cost, thereby constructing a power grid flood prevention risk probability prediction model based on the improved Transformer, and realizing high-efficiency and accurate calculation of the power grid flood prevention risk probability.
The invention also provides a power grid flood prevention risk prediction system based on the improved Transformer, which comprises the following steps: the system comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the computer program to realize any grid flood prevention risk prediction method based on the improved Transformer.
The processor and the memory may be connected by a bus or other means.
The memory, as a non-transitory computer readable storage medium, may be used to store non-transitory software programs as well as non-transitory computer executable programs. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory optionally includes memory remotely located relative to the processor, the remote memory being connectable to the processor through a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
It should be noted that, the road scene target detection system based on the optimized network structure and the mixed confidence coefficient in the embodiment may include a service processing module, an edge database, a server version information register, and a data synchronization module, where the processor implements the improved transform-based power grid flood prevention risk probability prediction method applied to the road scene target detection system based on the optimized network structure and the mixed confidence coefficient when executing the computer program.
The above described apparatus embodiments are merely illustrative, wherein the units illustrated as separate components may or may not be physically separate, i.e. may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
Furthermore, the present invention provides a computer readable storage medium storing computer executable instructions, where the computer executable instructions are executed by a processor or a controller, for example, by one of the processors in the above terminal embodiment, and cause the processor to execute the grid flood prevention risk prediction method based on the improved converter in the above embodiment.
Those of ordinary skill in the art will appreciate that all or some of the steps, systems, and methods disclosed above may be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on computer readable media, which may include computer storage media (or non-transitory media) and communication media (or transitory media). The term computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data, as known to those skilled in the art. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital Versatile Disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by a computer. Furthermore, as is well known to those of ordinary skill in the art, communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media.
While the preferred embodiment of the present invention has been described in detail, the present invention is not limited to the above embodiment, and various equivalent modifications and substitutions can be made by those skilled in the art without departing from the spirit of the present invention, and these equivalent modifications and substitutions are intended to be included in the scope of the present invention as defined in the appended claims.
The above-described embodiments of the present invention do not limit the scope of the present invention. Any other corresponding changes and modifications made in accordance with the technical idea of the present invention shall be included in the scope of the claims of the present invention.
Claims (9)
1. The power grid flood prevention risk prediction method based on the improved Transformer is characterized by comprising the following steps of:
acquiring flood prevention data;
carrying out power grid flood prevention prediction early warning influence factor identification analysis on flood prevention data;
modifying the Transformer network framework using a feature enhancement modification strategy based on a gating selection mechanism and embedded coding;
then, an explicit sparse attention improvement strategy is used for improving a Transformer network framework;
and carrying out power grid flood prevention risk prediction by using the improved converter network framework.
2. The improved fransformer-based power grid flood prevention risk probability prediction method according to claim 1, wherein before the power grid flood prevention prediction early warning impact factor identification analysis is performed on the flood prevention data, the method further comprises:
after flood control data are acquired, data preprocessing is carried out, and then identification analysis of power grid flood control prediction early warning influence factors is carried out;
the data preprocessing comprises data cleaning and data normalization;
the data cleaning is used for removing abnormal values in the data set and filling missing values;
the data normalization is used to make the data features of the same metric.
3. The improved Transformer-based power grid flood prevention risk probability prediction method of claim 1, wherein the performing power grid flood prevention prediction early warning impact factor identification analysis on the data comprises:
dividing monitoring information of a transformer substation into meteorological monitoring information, geographical environment information and site self information;
the information of the geographic environment and the site is the static covariate characteristic of the transformer substation;
the weather monitoring information is a dynamic covariate characteristic.
4. The improved Transformer based grid flood prevention risk probability prediction method of claim 1, wherein the feature enhancement improvement strategy using the gating selection mechanism and embedded coding comprises: the main body input feature selection strategy based on the gating mechanism and the time stamp feature fusion strategy based on the embedded coding.
5. The improved Transformer-based grid flood prevention risk probability prediction method of claim 4, wherein the gating mechanism-based principal input feature selection policy comprises:
taking the core meteorological features as main body input feature vectors, taking the static covariate features of the site as optional context feature vectors, and taking the main body input feature vectors and the optional context feature vectors as input features;
the input characteristics are operated through a gating residual error network module and softmax to obtain the selection weight of the input variable;
performing weighted assignment on the selection weight of the input variable and the input feature to obtain example feature selection of the input feature;
a feature variable selection network is constructed for the static covariates and time-dependent sequence variables applied to the instance feature selection.
6. The improved Transformer-based grid flood prevention risk probability prediction method of claim 4, wherein the embedded coding-based time stamp feature fusion strategy comprises:
s1, taking an input sequence with a reserved t moment as an input transformation sequence vector;
s2, embedding a local position feature vector of a storage sequence by using a fixed position;
s3, representing the global timestamp feature vector by using a learnable time code;
s4, mapping the input transformation sequence, the local position feature and the global time stamp feature vector to the same feature dimension by using a one-dimensional convolution filter with the kernel size of 3 and the step distance of 1, and fusing to obtain the enhanced input feature embedded code.
7. The improved Transformer-based grid flood prevention risk probability prediction method of claim 1, wherein the using explicit sparse attention improvement strategy is implemented by employing top-k selection strategy, comprising:
obtaining an original attention score matrix by using keys in a self-attention mechanism and query calculation;
multiplying the original attention score matrix with a threshold matrix, and then calculating by a sign function to obtain an original position mask matrix;
replacing 1 in the original position mask matrix with 0 and replacing 0 with minus infinity to obtain a position mask matrix;
and adding the original attention score matrix and the position mask matrix to obtain an attention score matrix.
8. An improved fransformer-based power grid flood prevention risk prediction system, comprising: a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the improved Transformer based grid flood prevention risk probability prediction method according to any one of claims 1-7 when the computer program is executed.
9. A computer-readable storage medium storing computer-executable instructions for causing a computer to perform the improved Transformer-based grid flood prevention risk probability prediction method according to any one of claims 1-7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310178028.3A CN116151459A (en) | 2023-02-28 | 2023-02-28 | Power grid flood prevention risk probability prediction method and system based on improved Transformer |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310178028.3A CN116151459A (en) | 2023-02-28 | 2023-02-28 | Power grid flood prevention risk probability prediction method and system based on improved Transformer |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116151459A true CN116151459A (en) | 2023-05-23 |
Family
ID=86373427
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310178028.3A Pending CN116151459A (en) | 2023-02-28 | 2023-02-28 | Power grid flood prevention risk probability prediction method and system based on improved Transformer |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116151459A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116738170A (en) * | 2023-06-13 | 2023-09-12 | 无锡物联网创新中心有限公司 | Abnormality analysis method and related device for industrial equipment |
CN117435918A (en) * | 2023-12-20 | 2024-01-23 | 杭州市特种设备检测研究院(杭州市特种设备应急处置中心) | Elevator risk early warning method based on spatial attention network and feature division |
Citations (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR102181966B1 (en) * | 2019-07-22 | 2020-11-23 | 치루 유니버시티 오브 테크놀로지 | Soft survey method and system for hydraulic cylinder comprehensive test station |
CN112734017A (en) * | 2021-04-01 | 2021-04-30 | 成都信息工程大学 | Method for establishing precipitation prediction model |
US20210293770A1 (en) * | 2017-06-19 | 2021-09-23 | Nanjing Institute Of Geography & Limnology. Chinese Academy Of Sciences | Stereoscopic monitoring and data mining system and method for harmful lake cyanobacteria bloom |
CN113435451A (en) * | 2021-06-28 | 2021-09-24 | 华为技术有限公司 | Model, training method and device of model, and recognition and device of character sequence |
CN114091361A (en) * | 2022-01-24 | 2022-02-25 | 中汽数据(天津)有限公司 | Weather event based transformer model construction method |
CN114238054A (en) * | 2021-12-17 | 2022-03-25 | 山东省计算中心(国家超级计算济南中心) | Cloud server resource utilization quantity prediction method based on improved TFT |
US20220108212A1 (en) * | 2020-10-01 | 2022-04-07 | Apple Inc. | Attention free transformer |
CN114503124A (en) * | 2019-12-18 | 2022-05-13 | 谷歌有限责任公司 | Multi-level prediction for processing time series data |
CN114756720A (en) * | 2022-06-13 | 2022-07-15 | 支付宝(杭州)信息技术有限公司 | Time sequence data prediction method and device |
CN115018553A (en) * | 2022-06-30 | 2022-09-06 | 东南大学 | Regional logistics single quantity prediction system and method based on deep learning |
CN115220133A (en) * | 2022-07-19 | 2022-10-21 | 海南大学 | Multi-meteorological-element rainfall prediction method, device, equipment and storage medium |
WO2022236064A2 (en) * | 2021-05-06 | 2022-11-10 | Strong Force Iot Portfolio 2016, Llc | Quantum, biological, computer vision, and neural network systems for industrial internet of things |
WO2022260432A1 (en) * | 2021-06-08 | 2022-12-15 | 네오사피엔스 주식회사 | Method and system for generating composite speech by using style tag expressed in natural language |
-
2023
- 2023-02-28 CN CN202310178028.3A patent/CN116151459A/en active Pending
Patent Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20210293770A1 (en) * | 2017-06-19 | 2021-09-23 | Nanjing Institute Of Geography & Limnology. Chinese Academy Of Sciences | Stereoscopic monitoring and data mining system and method for harmful lake cyanobacteria bloom |
KR102181966B1 (en) * | 2019-07-22 | 2020-11-23 | 치루 유니버시티 오브 테크놀로지 | Soft survey method and system for hydraulic cylinder comprehensive test station |
US20230018125A1 (en) * | 2019-12-18 | 2023-01-19 | Google Llc | Processing Multi-Horizon Forecasts For Time Series Data |
CN114503124A (en) * | 2019-12-18 | 2022-05-13 | 谷歌有限责任公司 | Multi-level prediction for processing time series data |
US20220108212A1 (en) * | 2020-10-01 | 2022-04-07 | Apple Inc. | Attention free transformer |
CN112734017A (en) * | 2021-04-01 | 2021-04-30 | 成都信息工程大学 | Method for establishing precipitation prediction model |
WO2022236064A2 (en) * | 2021-05-06 | 2022-11-10 | Strong Force Iot Portfolio 2016, Llc | Quantum, biological, computer vision, and neural network systems for industrial internet of things |
WO2022260432A1 (en) * | 2021-06-08 | 2022-12-15 | 네오사피엔스 주식회사 | Method and system for generating composite speech by using style tag expressed in natural language |
CN113435451A (en) * | 2021-06-28 | 2021-09-24 | 华为技术有限公司 | Model, training method and device of model, and recognition and device of character sequence |
CN114238054A (en) * | 2021-12-17 | 2022-03-25 | 山东省计算中心(国家超级计算济南中心) | Cloud server resource utilization quantity prediction method based on improved TFT |
CN114091361A (en) * | 2022-01-24 | 2022-02-25 | 中汽数据(天津)有限公司 | Weather event based transformer model construction method |
CN114756720A (en) * | 2022-06-13 | 2022-07-15 | 支付宝(杭州)信息技术有限公司 | Time sequence data prediction method and device |
CN115018553A (en) * | 2022-06-30 | 2022-09-06 | 东南大学 | Regional logistics single quantity prediction system and method based on deep learning |
CN115220133A (en) * | 2022-07-19 | 2022-10-21 | 海南大学 | Multi-meteorological-element rainfall prediction method, device, equipment and storage medium |
Non-Patent Citations (2)
Title |
---|
丁鑫: ""基于深度学习的西北太平洋热带气旋强度确定及预测研究"", 《中国优秀硕士学位论文全文数据库基础科学辑》, no. 10, 15 October 2022 (2022-10-15), pages 1 - 56 * |
上杉翔⼆: ""Transformer变体(Sparse Transformer, Longformer,Switch Transformer)"", 《CSDN》, 1 February 2021 (2021-02-01), pages 1 - 8 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116738170A (en) * | 2023-06-13 | 2023-09-12 | 无锡物联网创新中心有限公司 | Abnormality analysis method and related device for industrial equipment |
CN117435918A (en) * | 2023-12-20 | 2024-01-23 | 杭州市特种设备检测研究院(杭州市特种设备应急处置中心) | Elevator risk early warning method based on spatial attention network and feature division |
CN117435918B (en) * | 2023-12-20 | 2024-03-15 | 杭州市特种设备检测研究院(杭州市特种设备应急处置中心) | Elevator risk early warning method based on spatial attention network and feature division |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111161535B (en) | Attention mechanism-based graph neural network traffic flow prediction method and system | |
CN107529651B (en) | Urban traffic passenger flow prediction method and equipment based on deep learning | |
CN111400620B (en) | User trajectory position prediction method based on space-time embedded Self-orientation | |
CN116151459A (en) | Power grid flood prevention risk probability prediction method and system based on improved Transformer | |
Li et al. | DCT-GAN: dilated convolutional transformer-based GAN for time series anomaly detection | |
CN111080032A (en) | Load prediction method based on Transformer structure | |
CN113673769B (en) | Traffic flow prediction method of graph neural network based on multivariate time sequence interpolation | |
CN115587454A (en) | Traffic flow long-term prediction method and system based on improved Transformer model | |
CN115223049B (en) | Knowledge distillation and quantification method for large model compression of electric power scene edge calculation | |
CN113762338B (en) | Traffic flow prediction method, equipment and medium based on multiple graph attention mechanism | |
CN111242351A (en) | Tropical cyclone track prediction method based on self-encoder and GRU neural network | |
CN117096867A (en) | Short-term power load prediction method, device, system and storage medium | |
CN114692762A (en) | Vehicle track prediction method based on graph attention interaction mechanism | |
CN116092294A (en) | Expressway traffic flow prediction method based on Transformer and graph meaning network | |
CN115618248A (en) | Load abnormity identification method and device for public building, storage medium and equipment | |
Tschuchnig et al. | Sequential iot data augmentation using generative adversarial networks | |
Blier-Wong et al. | Geographic ratemaking with spatial embeddings | |
CN115049919A (en) | Attention regulation based remote sensing image semantic segmentation method and system | |
CN117132089B (en) | Power utilization strategy optimization scheduling method and device | |
CN115392595B (en) | Time-space short-term wind speed prediction method and system based on graph convolution neural network and Transformer | |
Senarathna et al. | An image based approach of energy signal disaggregation using artificial intelligence | |
CN115796359A (en) | PM2.5 space-time prediction method based on depth Koopman operator | |
CN114707829A (en) | Target person rescission risk prediction method based on structured data linear expansion | |
Bosma et al. | Estimating solar and wind power production using computer vision deep learning techniques on weather maps | |
CN114997490A (en) | Construction method, prediction method, device and equipment of temperature profile prediction model |
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 |