[go: up one dir, main page]

CN113222722A - Power budget assessment method based on door structure depth crossing network - Google Patents

Power budget assessment method based on door structure depth crossing network Download PDF

Info

Publication number
CN113222722A
CN113222722A CN202110514430.5A CN202110514430A CN113222722A CN 113222722 A CN113222722 A CN 113222722A CN 202110514430 A CN202110514430 A CN 202110514430A CN 113222722 A CN113222722 A CN 113222722A
Authority
CN
China
Prior art keywords
network
power budget
features
model
crossing
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110514430.5A
Other languages
Chinese (zh)
Other versions
CN113222722B (en
Inventor
许斌锋
仲田
王青国
胡扬波
陆野
徐进
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiangsu Electric Power Information Technology Co Ltd
Original Assignee
Jiangsu Electric Power Information Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Jiangsu Electric Power Information Technology Co Ltd filed Critical Jiangsu Electric Power Information Technology Co Ltd
Priority to CN202110514430.5A priority Critical patent/CN113222722B/en
Publication of CN113222722A publication Critical patent/CN113222722A/en
Application granted granted Critical
Publication of CN113222722B publication Critical patent/CN113222722B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/12—Accounting
    • G06Q40/125—Finance or payroll
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
    • G06Q10/06313—Resource planning in a project environment
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
    • Y02E40/70—Smart grids as climate change mitigation technology in the energy generation sector
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

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

Abstract

The invention discloses a power budget assessment method based on a door structure deep crossing network, which starts from 2 dimensions of a 'first-level profit center' and a 'business domain', and extracts representative characteristics influencing power budget; the Embedding network layer is adopted to convert the characteristics from discrete and continuous types into low-dimensional dense characterization vectors respectively; and capturing interaction among different features by adopting a multi-head attention network in an open-source Transformer model and learning the ambiguity brought by diversified feature interaction, thereby outputting a high-quality feature expression vector. Respectively transmitting the characterization vectors into a depth network and a cross network for fusion; and calculating the power budget assessment value of the future year of the sample in the test set based on the trained model. The invention solves the problem that the annual budget and distribution process of power system units and professional departments is lack of decision guidance, not only maintains the advantages of a deep network, but also can utilize a cross network to perform explicit cross calculation on characteristics.

Description

Power budget assessment method based on door structure depth crossing network
Technical Field
The invention belongs to the field of electric power, and particularly relates to an electric power budget assessment method based on a door structure depth crossing network.
Background
With the rapid development of national economy, in order to meet the actual development demand of society, power companies need to make intelligent innovation and reform from the aspects of internal system and development strategy. The electric power financial budget assessment is an important component in the management work of electric power enterprises, in the actual development and operation process, the electric power enterprises can effectively solve the problems existing in the electric power financial budget management only by fully paying attention to the electric power financial budget management work and combining the actual development condition, and the electric power financial budget assessment fundamentally provides electric power guarantee for the development of national economy in China. When the existing electric power company performs pre-calculation distribution on a first-level profit center and a business domain, the problem of taking experience as a main part exists, a set of complete scientific system structure and model are lacked to ensure the reasonability and feasibility of budget distribution, and related technical support is lacked.
At present, the existing research work aiming at power budget prediction mainly focuses on extracting relevant characteristic basic knowledge, a machine learning algorithm is adopted as a main part, and a prediction model is established by combining socioeconomic data such as power budget, city GDP, city population and the like in the past so as to improve the scientificity of budget allocation, such as a linear regression model, a logistic regression model, a decision tree model, a deep network, a cross network and the like. The above models often ignore the synergistic effect between features and also fail to eliminate semantic ambiguity between features. In addition, in the aspect of vector fusion of deep network learning and cross network learning, the difference of the importance of the 2 types of network learning vectors is also ignored.
Disclosure of Invention
The invention aims to provide a door structure depth crossing network-based power budget assessment method, which is used for solving the unscientific and unreasonable problems in the existing power budget allocation, so that a new idea is developed for a new method and a new mode for overall planning and cultivating financial management of various professional resources, and the door structure depth crossing network-based power budget assessment method can be applied to different power financial prediction scenes.
In order to solve the problems, the technical scheme adopted by the invention is as follows:
a power budget assessment method based on a gate structure depth crossing network comprises the following steps of firstly, carrying out preprocessing related operation on dimensional data of a first-level profit center of an enterprise and dimensional data of a service domain; secondly, starting from 2 dimensions of a 'first-level profit center' and a 'business domain', 13 representative features influencing the electric power budget are extracted from the 2 types of data respectively; thirdly, an Embedding network layer is adopted to convert the features from discrete types and continuous types into low-dimensional dense characterization vectors respectively; thirdly, a multi-head attention network of the Google Transformer model is adopted to capture interaction between different features and learn the ambiguity brought by diversified feature interaction, so that high-quality feature expression vectors are output. Then, the characterization vectors are respectively transmitted into a deep network and a cross network, the deep network is used for implicitly generating interaction and the cross network among the features to perform explicit cross calculation on the features, the workload of manually performing feature cross is reduced, and the vectors learned by the 2 types of networks are fused through a gate structure network; and finally, calculating the power budget assessment value of the future year of the sample in the test set based on the trained model.
The method comprises the following steps:
1) and analyzing the power budget distribution data of the past year from 2 dimensions of the primary profit center and the service domain. And excavating influence characteristics influencing power budget distribution data of the past year, wherein the influence characteristics comprise a first-level profit center dimension data characteristic and a business domain dimension data characteristic 2 part, preprocessing relevant operations are carried out on the characteristics, and a training set and a test set are divided.
2) Establishing a power budget evaluation model based on a gate structure deep cross network, implicitly generating interaction and cross networks among the features by using the deep network to perform explicit cross calculation on the features, and fusing vectors learned by the 2 types of networks through the gate structure network. Inputting all the characteristic values in the training set obtained in the step 1) into the method for training and learning to obtain high-order characteristic expression vectors, and establishing a non-linear relation between the power budget allocation value and the characteristic values in the past year.
3) Optimizing the power budget assessment model based on the door structure depth crossing network designed in the step 2) in a training phase by means of an Adam optimizer to minimize a loss function so as to obtain optimal parameters.
4) And 3) according to the optimized model obtained in the step 3), in a testing stage, finally, the automatic generation of the electric power budget proposal scheme is realized by inputting relevant characteristic values.
The specific method for analyzing the power budget allocation data and the influence factors thereof in the step 1) over the years comprises the following steps: and analyzing the dimensional data of the primary profit center of the power enterprise, and finding out parameters with high correlation with the power budget distribution data of the past year, such as the number of staff of each power supply company, the number of equipment, the number of substations of each voltage class, the number of power transmission lines, the power selling amount, the power transmission amount, the original value of assets, the standard operation cost (unit price) and the like. And analyzing the service domain dimension data, and finding out parameters with high correlation degree with the power budget allocation in the past year, such as city GDP, population and other economic indexes. And selecting reasonable features with data in all years from the selected features, analyzing the data value or accumulated value of the feature in the current year, and selecting a proper value for each feature, wherein the features are classified into a discrete type and a continuous type and are input into a deep learning model. The data set is partitioned in a ratio of training set to test set 9: 1.
The specific operation of training and learning by using the power budget evaluation model based on the door structure depth crossing network in the step 2) is as follows:
first, the class 2 feature set group vector is added
Figure 718136DEST_PATH_IMAGE001
And
Figure 900856DEST_PATH_IMAGE002
inputting two Embedding network layers, and respectively converting the features from discrete and continuous types into low-dimensional dense characterization vectors which are recorded as
Figure 228063DEST_PATH_IMAGE003
Finally, the learned class 2 vectors are spliced by means of the Concat function in the neural network and are recorded as
Figure 154431DEST_PATH_IMAGE004
。
Secondly, the influence of the cooperative influence between the features and the feature interaction of conflict semantics on the prediction result is considered. The invention adopts the multi-head attention network of the Google Transformer model to capture the interaction between different characteristics and the ambiguity brought by learning diversified characteristic interaction, and meanwhile, the model has strong parallel computability and can efficiently output high-quality characteristic expression vectors. Given the input vector of the Transformer model
Figure 703224DEST_PATH_IMAGE006
(iv) Transformer of
Figure 56845DEST_PATH_IMAGE008
Latent expression vector of individual head
Figure 120616DEST_PATH_IMAGE010
This can be found by scaling the Dot-Product Attention (Dot-Product Attention):
Figure 621914DEST_PATH_IMAGE011
wherein Q, K, V represents the three vectors of Query, Key and Value in the transform model respectively,
Figure 25214DEST_PATH_IMAGE012
and
Figure 549736DEST_PATH_IMAGE013
is used for learning Transformer
Figure 100803DEST_PATH_IMAGE014
The weight parameters of the individual heads are set,
Figure 368973DEST_PATH_IMAGE016
is the dimension of the vector K. Hidden features
Figure 626779DEST_PATH_IMAGE017
Form an enhanced token vector
Figure 338515DEST_PATH_IMAGE018
Information inherent to each feature and ambiguous information are stored. The present invention combines a feedforward data network with an excitation function to learn a nonlinear combination of information:
Figure 376878DEST_PATH_IMAGE019
wherein,
Figure 183160DEST_PATH_IMAGE020
is a weight that can be trained in a way that,
Figure 561052DEST_PATH_IMAGE021
is the number of attention heads; representing the concatenation of the vectors.
Thirdly, the feature vector of the feature is measured
Figure 427376DEST_PATH_IMAGE022
The method is respectively input into a deep network and a cross network, aims to implicitly generate the interaction between the features by using the deep network, and simultaneously performs explicit cross calculation on the features by using the cross network, so that the workload of manually performing feature cross is reduced. In particular, Cross networks (Cross networks) are composed of multiple networksThe calculation mode of the first layer is as follows:
Figure 703768DEST_PATH_IMAGE023
the calculation method of each cross layer after the second layer is as follows:
Figure 313741DEST_PATH_IMAGE024
here, the number of the first and second electrodes,
Figure 546139DEST_PATH_IMAGE025
and
Figure 848945DEST_PATH_IMAGE026
respectively represent
Figure 596321DEST_PATH_IMAGE027
Layer and the first
Figure 760717DEST_PATH_IMAGE028
The output vector of the layer-crossing network,
Figure 644359DEST_PATH_IMAGE029
and
Figure 321328DEST_PATH_IMAGE030
are respectively the first
Figure 290421DEST_PATH_IMAGE027
Weights and biases of the layer crossing network.
The Deep Network (Deep Network) is a fully-connected feedforward neural Network, and the calculation mode of the first layer is as follows:
Figure 507776DEST_PATH_IMAGE031
the calculation method of each deep network after the second layer is as follows:
Figure 259307DEST_PATH_IMAGE032
wherein,
Figure 903915DEST_PATH_IMAGE033
and
Figure 563566DEST_PATH_IMAGE034
are respectively the first
Figure 319033DEST_PATH_IMAGE027
Layer and the first
Figure 177267DEST_PATH_IMAGE028
The output vector of the deep network of layers,
Figure 196039DEST_PATH_IMAGE035
and
Figure 890456DEST_PATH_IMAGE036
are respectively the first
Figure 715193DEST_PATH_IMAGE027
Weights and biases of the layer depth network. The total number of layers of the deep network and the cross network are allnAnd finally obtained expression vectors of the features are respectively recorded as
Figure 427934DEST_PATH_IMAGE037
And
Figure 352028DEST_PATH_IMAGE038
。
finally, the vector of the class 2 network learning is measured through a gate structure network
Figure 783009DEST_PATH_IMAGE037
And
Figure 896590DEST_PATH_IMAGE038
the importance of the method is fused, and the calculation mode is as follows:
Figure 729417DEST_PATH_IMAGE039
wherein,
Figure 824412DEST_PATH_IMAGE040
、
Figure 742689DEST_PATH_IMAGE041
and
Figure 643649DEST_PATH_IMAGE042
is a parameter in the door architecture convergence network,
Figure 81715DEST_PATH_IMAGE043
the excitation function chosen for the present invention. Finally, the pass-gate structure pair
Figure 409928DEST_PATH_IMAGE037
And
Figure 753184DEST_PATH_IMAGE038
the final characterization vector can be obtained by fusion
Figure 723414DEST_PATH_IMAGE044
The calculation method is as follows:
Figure 265254DEST_PATH_IMAGE045
finally, the process is carried out in a batch,
Figure 967631DEST_PATH_IMAGE046
representing the Hadamard product of two vectors. Use of
Figure 342724DEST_PATH_IMAGE047
As a function of excitation, will
Figure 851066DEST_PATH_IMAGE044
Conversion to predicted power budget:
Figure 247412DEST_PATH_IMAGE048
the sample loss function in the step 3) is a square loss function, and a training set is given
Figure 120690DEST_PATH_IMAGE049
，tIs a set
Figure 235277DEST_PATH_IMAGE049
The label (true power budget value) corresponding to a certain sample in (1) is
Figure 32463DEST_PATH_IMAGE050
The final square loss function is defined as follows:
Figure 752157DEST_PATH_IMAGE051
(
Figure 593074DEST_PATH_IMAGE052
the present invention utilizes an Adam optimizer to minimize the above loss function, thereby tuning the parameters in the predictive model to an optimal configuration.
In the step 4), by means of the power budget evaluation model trained in the step 3) and based on the door structure depth crossing network, the power budget value is finally output by inputting the relevant characteristic values in the test set, so that the automatic generation of the power budget proposal scheme is realized.
According to the invention, a multi-head attention network of a Google transform model is adopted to capture interaction between different features and ambiguity brought by learning diversified feature interaction, on the basis, a door structure deep crossing network is designed, the interaction and crossing network between the features are implicitly generated by the deep network to perform explicit crossing calculation on the features, the workload of manually performing feature crossing is reduced, vectors learned by the 2 types of networks are fused through the door structure network, and finally high-precision electric power financial budget assessment is realized.
The invention has the beneficial effects that: analyzing the power budget distribution data of the past year and corresponding related characteristics thereof from two dimensions of a first-level profit center and a service domain, designing a model based on a gate structure depth crossing network, and predicting the power budget distribution data of the future year.
In particular, the present invention has the following advantages:
1. the model based on the door structure deep crossing network can capture interaction among different features and ambiguity brought by learning diversified feature interaction, on the basis, the interaction and the crossing network among the features are implicitly generated by the deep network to perform explicit crossing calculation on the features, so that the workload of manually performing feature crossing is reduced, vectors learned by the 2 types of networks are fused through the door structure network, and finally high-precision electric power financial budget assessment is realized.
2. The model based on the door structure deep crossing network can predict the power budget allocation data in the coming year, solves the problems of lack of scientificity, urgent need of technical support and the like in the process of annual budget and allocation of power system units and professional departments, and can effectively improve the high efficiency and scientificity of resource allocation.
Drawings
FIG. 1 is an overall system framework of the present invention.
Detailed Description
The present invention is further illustrated by the following examples, which are intended to be purely exemplary and are not intended to limit the scope of the invention, which is defined in the appended claims, as interpreted by those skilled in the art.
Referring to fig. 1, the standard of deep learning is to use the same shape to represent vectors, different colors to represent different vectors, Concat and ReLu are commonly used function names, and Transformer is a model name. The invention discloses an automatic generation method of a power budget proposal scheme based on big data analysis, which comprises the following steps:
1) and analyzing the data of the power budget of the past year from 2 dimensions of the first-level profit center and the business domain. And excavating influence characteristics influencing power budget distribution data of the past year, wherein the influence characteristics comprise a primary profit center dimension data characteristic and a service domain dimension data characteristic 2 part, carrying out preprocessing related operation on the characteristics, and dividing a data set according to the proportion of a training set to a test set 9: 1.
2) And establishing a power budget assessment model based on a door structure depth crossing network. Inputting all the characteristics in the training set obtained in the step 2) into the model for training and learning to obtain high-order characteristic expression vectors, and establishing a nonlinear relation between the power budget allocation value and the characteristic value in the past year.
3) Optimizing the power budget assessment based on the gate structure depth crossing network designed in the step 3) by means of an Adam optimizer to minimize a loss function so as to obtain an optimal parameter configuration.
4) And finally, automatically generating the electric power budget proposal scheme by inputting relevant characteristic values according to the optimized model obtained in the step 3).
The specific method for analyzing the power budget allocation data and the influence factors thereof in the step 1) over the years comprises the following steps: analyzing the dimensional data of the primary profit center of the power enterprise, finding out the characteristic with higher correlation degree with the power budget allocation data of the past year, analyzing the dimensional data of the service domain, finding out the characteristic with higher correlation degree with the power budget allocation data of the past year, and finally extracting the characteristics as shown in table 1. Selecting reasonable characteristics with data in all years from the selected characteristics, analyzing the data value or accumulated value of the characteristics in the current year, selecting proper value for each characteristic, classifying the characteristics in the sets into discrete type and continuous type, and recording the set formed by the 2 types of characteristics asF
Figure 194957DEST_PATH_IMAGE053
The vector is the input to the deep learning model. The present invention partitions the data set according to the ratio of the training set to the test set 9: 1.
Table 1 data characterization
Figure 45101DEST_PATH_IMAGE054
The specific operation of training and learning by using the power budget assessment method based on the gate structure depth crossing network in the step 2) is as follows:
first, the class 2 feature set group vector is added
Figure 166772DEST_PATH_IMAGE001
And
Figure 381853DEST_PATH_IMAGE002
inputting two Embedding network layers, and respectively converting the features from discrete and continuous types into low-dimensional dense characterization vectors which are recorded as
Figure 205452DEST_PATH_IMAGE003
Finally, the learned class 2 vectors are spliced by means of the Concat function in the neural network and are recorded as
Figure 859287DEST_PATH_IMAGE055
. Further, the influence of the cooperative influence between the features and the feature interaction of conflict semantics on the prediction result is considered. The invention adopts the multi-head attention network of the Google Transformer model to capture the interaction between different characteristics and the ambiguity brought by the characteristic interaction of learning diversity, and meanwhile, the model has strong parallel computability, can efficiently output high-quality characteristic expression vectors and is recorded as the characteristic expression vectors
Figure 84732DEST_PATH_IMAGE056
。
Thirdly, the feature vector of the feature is measured
Figure 18184DEST_PATH_IMAGE056
The method is respectively input into a deep network and a cross network, and aims to implicitly generate the interaction between the features by using the deep network and simultaneously perform explicit cross calculation on the features by using the cross network. Specifically, a Cross Network (Cross Network) is composed of a plurality of Cross layers, and the calculation method of the first layer is as follows:
Figure 797921DEST_PATH_IMAGE057
the calculation method of each cross layer after the second layer is as follows:
Figure 989868DEST_PATH_IMAGE058
,
here, the number of the first and second electrodes,
Figure 69820DEST_PATH_IMAGE059
and
Figure 892282DEST_PATH_IMAGE060
respectively representlLayer and the firstlThe output vector of the +1 layer cross network,
Figure 690474DEST_PATH_IMAGE061
and
Figure 168335DEST_PATH_IMAGE062
are respectively the firstlWeights and biases of the layer crossing network.
The Deep Network (Deep Network) is a fully-connected feedforward neural Network, and the calculation mode of the first layer is as follows:
Figure 368373DEST_PATH_IMAGE063
the calculation method of each deep network after the second layer is as follows:
Figure 361736DEST_PATH_IMAGE064
wherein,
Figure 647224DEST_PATH_IMAGE065
and
Figure 180974DEST_PATH_IMAGE066
are respectively the firstlLayer and the firstlThe output vector of the deep network of layer +1,
Figure 986250DEST_PATH_IMAGE067
and
Figure 150515DEST_PATH_IMAGE068
are respectively the firstlWeights and biases of the layer depth network. The total number of layers of the deep network and the cross network are allnAnd finally obtained expression vectors of the features are respectively recorded as
Figure 657720DEST_PATH_IMAGE037
And
Figure 729581DEST_PATH_IMAGE038
。
finally, a gate structure network is designed to measure the learning vector of the 2 types of networks
Figure 169789DEST_PATH_IMAGE037
And
Figure 786847DEST_PATH_IMAGE038
the importance of the method is fused, and the calculation mode is as follows:
Figure 250189DEST_PATH_IMAGE069
wherein,
Figure 860162DEST_PATH_IMAGE040
、
Figure 154877DEST_PATH_IMAGE041
and
Figure 192103DEST_PATH_IMAGE042
is a parameter in the door architecture convergence network,
Figure 690212DEST_PATH_IMAGE043
the excitation function chosen for the present invention. Finally, the pass-gate structure pair
Figure 307138DEST_PATH_IMAGE037
And
Figure 190780DEST_PATH_IMAGE038
the final characterization vector can be obtained by fusion
Figure 930066DEST_PATH_IMAGE044
The calculation method is as follows:
Figure 899159DEST_PATH_IMAGE071
finally, use
Figure 54197DEST_PATH_IMAGE047
As a function of excitation, will
Figure 805728DEST_PATH_IMAGE044
Conversion to predicted power budget:
Figure 184757DEST_PATH_IMAGE048
the sample loss function in the step 3) is a square loss function, and a training set is given
Figure 906725DEST_PATH_IMAGE049
，tIs a set
Figure 662191DEST_PATH_IMAGE049
The label (true power budget value) corresponding to a certain sample in (1) is
Figure 723688DEST_PATH_IMAGE050
The final square loss function is defined as follows:
Figure 289930DEST_PATH_IMAGE051
(
Figure 499194DEST_PATH_IMAGE052
the present invention utilizes an Adam optimizer to minimize the above loss function, thereby tuning the parameters in the predictive model to an optimal configuration.
In the step 4), by means of the power budget evaluation model trained in the step 3) and based on the door structure depth crossing network, the power budget value is finally output by inputting the relevant characteristic values in the test set, so that the automatic generation of the power budget proposal scheme is realized.

Claims (5)

1. A power budget assessment method based on a gate structure depth crossing network is characterized in that: firstly, starting from 2 dimensions of a 'primary profit center' and a 'business domain', respectively extracting representative characteristics influencing power budget from data of the 2 dimensions; thirdly, an Embedding network layer is adopted to convert the features from discrete types and continuous types into low-dimensional dense characterization vectors respectively; then, splicing the characterization vectors, and transmitting the spliced characterization vectors into a model based on a gate structure depth crossing network; finally, training the model to obtain the optimal parameter configuration, and calculating the power budget assessment value of the test set sample in the future year, specifically comprising the following steps:
1) analyzing the power budget distribution data of the past year from 2 dimensions of the primary profit center and the service domain, mining influence characteristics influencing the power budget distribution data of the past year, wherein the influence characteristics comprise a primary profit center dimension data characteristic and a service domain dimension data characteristic 2 part, preprocessing relevant operations are carried out on the characteristics, and a training set and a test set are divided;
2) establishing a power budget evaluation model based on a gate structure deep crossing network, inputting all the characteristics in the training set obtained in the step 1) into the model for training and learning to obtain a high-order characteristic expression vector, and establishing a nonlinear relation between a power budget allocation value and characteristic values in the past year;
3) in a training stage, optimizing the power budget evaluation model based on the door structure depth crossing network in the step 2), and minimizing a loss function by means of an Adam optimizer to obtain optimal parameters;
4) and 3) according to the optimized power budget assessment model based on the door structure depth crossing network obtained in the step 3), in a testing stage, finally, automatic generation of a power budget proposal scheme is realized by inputting relevant characteristic values.
2. The method for estimating the power budget based on the gate structure deep crossing network according to claim 1, wherein in the step 1), starting from 2 dimensions of a 'primary profit center' and a 'business domain', features influencing the power budget are extracted from the data of the 2 dimensions respectively, the features which have data and are reasonable over the years are selected, the data value or the accumulated value of the year is analyzed, and a proper value is selected for each feature, wherein the features are classified into a discrete type and a continuous type and are input into a deep learning model; the data set is partitioned in a ratio of training set to test set 9: 1.
3. The method for estimating power budget based on the gate structure depth crossing network as claimed in claim 1, wherein the Embedding network layer is adopted in step 2) to convert the features from discrete type and continuous type into low-dimensional dense feature vectors respectively; then, splicing the characterization vectors, and transmitting the spliced characterization vectors into a model based on a gate structure depth crossing network; and finally, in the training stage, training the model to obtain optimal parameters, and calculating the power budget assessment value of the future year of the sample in the test set.
4. The door structure depth crossing network-based power budget assessment method according to claim 3, wherein a multi-head attention network of a Google Transformer model is adopted to capture interaction between different features and learn ambiguity brought by diversified feature interaction, a door structure depth crossing network-based model is established, the interaction between the features and the crossing network are implicitly generated by using the depth network to perform explicit crossing calculation on the features, and vectors learned by the 2 types of networks are fused by the door structure network, so that high-precision power financial budget assessment is finally achieved.
5. The method for estimating power budget based on the door structure depth crossing network according to claim 1, wherein in the step 3), the interaction between the features is implicitly generated by using the depth network, and meanwhile, the features are explicitly crossed and calculated by using the crossing network, so that the workload of manually crossing the features is reduced, and in a model training stage, an Adam optimizer is used for minimizing a square loss function, thereby adjusting parameters in a prediction model to be in an optimal configuration.
CN202110514430.5A 2021-05-12 2021-05-12 Power budget assessment method based on gate structure depth crossing network Active CN113222722B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110514430.5A CN113222722B (en) 2021-05-12 2021-05-12 Power budget assessment method based on gate structure depth crossing network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110514430.5A CN113222722B (en) 2021-05-12 2021-05-12 Power budget assessment method based on gate structure depth crossing network

Publications (2)

Publication Number Publication Date
CN113222722A true CN113222722A (en) 2021-08-06
CN113222722B CN113222722B (en) 2024-02-02

Family

ID=77095130

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110514430.5A Active CN113222722B (en) 2021-05-12 2021-05-12 Power budget assessment method based on gate structure depth crossing network

Country Status (1)

Country Link
CN (1) CN113222722B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115600146A (en) * 2022-10-24 2023-01-13 南通耀恒新能源有限公司（Cn） Control method of distributed photovoltaic grid-connected power generation and energy storage equipment based on big data
CN118425087A (en) * 2024-05-08 2024-08-02 电子科技大学 A coal quality component detection method and system based on deep learning

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8280805B1 (en) * 2006-01-10 2012-10-02 Sas Institute Inc. Computer-implemented risk evaluation systems and methods
CN109829587A (en) * 2019-02-12 2019-05-31 国网山东省电力公司电力科学研究院 Zonule grade ultra-short term and method for visualizing based on depth LSTM network
CN112348352A (en) * 2020-11-05 2021-02-09 国网江苏省电力有限公司 Automatic generation method of power budget proposal scheme based on big data analysis

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8280805B1 (en) * 2006-01-10 2012-10-02 Sas Institute Inc. Computer-implemented risk evaluation systems and methods
CN109829587A (en) * 2019-02-12 2019-05-31 国网山东省电力公司电力科学研究院 Zonule grade ultra-short term and method for visualizing based on depth LSTM network
CN112348352A (en) * 2020-11-05 2021-02-09 国网江苏省电力有限公司 Automatic generation method of power budget proposal scheme based on big data analysis

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
王恩创;任玉珑;朱春波;: "基于模糊DEA的配电网投入产出综合效率评估", 工业工程与管理, no. 02 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115600146A (en) * 2022-10-24 2023-01-13 南通耀恒新能源有限公司（Cn） Control method of distributed photovoltaic grid-connected power generation and energy storage equipment based on big data
CN115600146B (en) * 2022-10-24 2025-07-29 南通耀恒新能源有限公司 Control method of distributed photovoltaic grid-connected power generation and energy storage equipment based on big data
CN118425087A (en) * 2024-05-08 2024-08-02 电子科技大学 A coal quality component detection method and system based on deep learning
CN118425087B (en) * 2024-05-08 2024-11-19 电子科技大学 Deep learning-based coal composition detection method and system

Also Published As

Publication number Publication date
CN113222722B (en) 2024-02-02

Similar Documents

Publication Publication Date Title
Zheng et al. A hybrid energy-aware resource allocation approach in cloud manufacturing environment
Chen An interval‐valued Pythagorean fuzzy outranking method with a closeness‐based assignment model for multiple criteria decision making
CN107480856A (en) Based on the sale of electricity company power customer appraisal procedure for improving similarity to ideal solution ranking method
CN109033497A (en) A kind of multistage data mining algorithm intelligent selecting method towards high concurrent
CN117934209B (en) Regional power system carbon emission big data analysis method based on knowledge graph
CN113222722B (en) Power budget assessment method based on gate structure depth crossing network
CN110688477B (en) A prediction method, device, equipment and storage medium in a technical hotspot field
Ramírez et al. Artificial intelligence and its impact on the prediction of economic indicators
CN112991032A (en) Power budget assessment method based on hierarchical attention network
CN112348352A (en) Automatic generation method of power budget proposal scheme based on big data analysis
Vokony et al. Agile digitalization evolution in the energy sector, taking into account innovative and disruptive technologies
Puech et al. Optimal Power Grid Operations with Foundation Models
CN115481861A (en) Power distribution network comprehensive benefit evaluation method and system
Zhao Predicting the Line Loss for a 10 kV Distribution Network Using AGA‐BPNN
CN113011779A (en) Energy consumption price compensation method and device based on fuzzy comprehensive evaluation
Zhao Implementation of Using AI to Manage Known and Unknown Risks in Risk Management
Mahmud et al. A Smart System for Monthly Electrical EnergyConsumption Prediction Using Machine Learning
CN108805366A (en) Multifactor adaptive neural network Methods of electric load forecasting based on decision formal context and system
Wang Introducing Cloud Data Technology into Economic Statistics
Chen [Retracted] The Prediction of English Online Network Performance Based on the XGBoost Algorithm
Zhang et al. A Novel LambdaRank-Based Method for Assessing the Cost-Effectiveness of Power Enterprise Projects
Liu et al. Monthly industrial added value monitoring model with multi-source big data
Wang et al. Research on line loss prediction model based on clustering and multi prediction model fusion
Yu et al. Design and Application of Personalized Recommendation Algorithm for Electricity Marketing
Yingxuan Research on the Construction System of Power Science and Technology Innovation Index and Evaluation System Based on Deep Learning

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant