CN112036649B - Hydropower station risk assessment method based on multi-core parallel runoff probability density prediction - Google Patents

Hydropower station risk assessment method based on multi-core parallel runoff probability density prediction Download PDF

Info

Publication number
CN112036649B
CN112036649B CN202010915902.3A CN202010915902A CN112036649B CN 112036649 B CN112036649 B CN 112036649B CN 202010915902 A CN202010915902 A CN 202010915902A CN 112036649 B CN112036649 B CN 112036649B
Authority
CN
China
Prior art keywords
runoff
formula
prediction
probability
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.)
Active
Application number
CN202010915902.3A
Other languages
Chinese (zh)
Other versions
CN112036649A (en
Inventor
何耀耀
张婉莹
陈悦
王云
肖经凌
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.)
Hefei University of Technology
Original Assignee
Hefei University of Technology
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 Hefei University of Technology filed Critical Hefei University of Technology
Priority to CN202010915902.3A priority Critical patent/CN112036649B/en
Publication of CN112036649A publication Critical patent/CN112036649A/en
Application granted granted Critical
Publication of CN112036649B publication Critical patent/CN112036649B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0635—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
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
    • H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • 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)
  • General Health & Medical Sciences (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Quality & Reliability (AREA)
  • Mathematical Physics (AREA)
  • Operations Research (AREA)
  • Game Theory and Decision Science (AREA)
  • Development Economics (AREA)
  • Power Engineering (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • Primary Health Care (AREA)
  • Educational Administration (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a hydropower station risk assessment method based on multi-core parallel runoff probability density prediction, which comprises the following steps: 1, acquiring runoff and related characteristic data and preprocessing the runoff and the related characteristic data; 2, arranging the preprocessed data set by using a rolling prediction method, and dividing the data set into a training set and a test set; 3, constructing an MPRVFL model of a multi-core parallel random vector function chain network, averagely dividing a training set, and then respectively carrying out parallel training on the model; substituting the test set into the trained MPRVFL model, and performing probability density prediction on the obtained conditional quantiles to obtain a prediction result and corresponding probability; and 5, grading the prediction results of the radial flow probability density, counting the number of water abandoning and load tasks which cannot be completed and calculating the corresponding risk probability. According to the method, idle resources of a computer are fully utilized, the running efficiency of the model is further improved while the runoff prediction precision is improved, and therefore decision basis can be provided for the medium-term and long-term hydrological forecast of the runoff.

Description

Hydropower station risk assessment method based on multi-core parallel runoff probability density prediction
Technical Field
The invention belongs to the field of hydropower station energy optimization, and particularly relates to a hydropower station risk assessment method based on multi-core parallel runoff probability density prediction.
Background
The utilization and development of water resources are beneficial to promoting the development of social economy, improving the energy consumption structure and slowing down the environmental pollution caused by the consumption of resources such as coal, petroleum and the like. Runoff prediction and hydropower station risk assessment are important issues in the development and utilization of water resources. Accurate and reliable runoff prediction and hydropower station risk assessment are effective means and key links for optimizing water resource distribution, realizing reasonable operation of a power grid and obtaining economic benefits. The runoff is influenced by natural conditions such as precipitation, landform and weather, and has the characteristics of obvious randomness and volatility, so that the runoff has extremely high certainty.
While runoff uncertainty is a major factor contributing to the risk of water abandonment and failure to complete the load task. Due to the fact that the accuracy of runoff prediction is limited, runoff prediction errors are large, load tasks of a power grid executed by a hydropower station are affected, and further decision deviation may be caused to cause unnecessary risks. Many hydropower stations usually face a serious water abandoning risk when encountering large flood in flood season, and can not meet the load requirement in the dry season. Therefore, runoff prediction is carried out by combining runoff related characteristic information, hydropower station risk assessment analysis is carried out, and a control strategy is reasonably formulated, so that the method is vital to reduce risks, reduce unnecessary water abandonment and complete load tasks.
The scale of basic data required to be collected by the conventional runoff prediction method is gradually increased, time is consumed for solving a gradually complex model, and the calculation cost becomes the constraint of runoff prediction. In addition, most runoff prediction models in the past are deterministic prediction models, only point prediction results of runoff can be obtained, and influence of uncertain factors on runoff fluctuation is difficult to reflect. The risk hidden in the deterministic prediction of runoff is not negligible. Therefore, realizing high-precision runoff and reducing risk reliable prediction caused by runoff prediction errors are theoretical and practical engineering problems which need to be solved urgently.
Disclosure of Invention
The invention aims to overcome the defects in the prior art, and provides a hydropower station risk assessment method based on multi-core parallel runoff probability density prediction, so that idle resources of a computer can be fully utilized, the running efficiency of a model is further improved while the runoff prediction precision is improved, the uncertainty and randomness of the runoff prediction can be fully considered, the hydropower station running risk can be efficiently and comprehensively assessed, and a decision basis is provided for long-term hydrological forecast in runoff.
In order to achieve the purpose, the invention adopts the following technical scheme:
the invention relates to a hydropower station risk assessment method based on multi-core parallel runoff probability density prediction, which is characterized by comprising the following steps of:
step 1, collecting runoff data at different time points and rainfall characteristic, air temperature characteristic and air pressure characteristic related to runoff, and normalizing to obtain a preprocessed runoff data set X ═ X' 1 ，x′ 2 ，…，x′ n ，…，x′ N ) The precipitation feature set P ═ P (P) 1 ，p 2 ，…，p n ，…，p N ) Temperature profile T ═ T 1 ，t 2 ，…，t n ，…，t N ) And set of air pressure characteristics R ═ R (R) 1 ，r 2 ，…，r n ，…，r N ) (ii) a Wherein x is n 、p n 、t n And r n Respectively representing runoff, precipitation, air temperature and air pressure at the ith time point after pretreatment, wherein N is 1,2, …, and N is data quantity acquired by each characteristic;
step 2, predicting the runoff data of the (M +1) th time point by using a rolling arrangement prediction method and respectively using the runoff data set X' and the data of the first M time points in each characteristic set related to runoff; thereby obtaining an (N-M) X (4M +1) dimensional rolling runoff matrix which is marked as (X,y); wherein X is (X) 1 ，X 2 ，…，X m ，…，X 4M ) Representing an input variable, X m Represents the m-th input variable, an
Figure BDA0002665003220000021
The jth sample representing the mth input variable, Y ═ Y 1 ，Y 2 ，…，Y m ，…，Y 4N-4M ) T Is an output variable, Y m Represents the mth sample in the output variable Y;
step 3, dividing the first l rows of the (N-M) X (4M +1) dimension rolling matrix (X, Y) into a training set (X) Tr ，Y Tr ) And the rest as test set (X) Te ，Y Te ) (ii) a Wherein l is more than or equal to 1 and less than 4N-4M;
step 4, training the training set (X) Tr ，Y Tr ) Divided equally into K subsets
Figure BDA0002665003220000022
Wherein the content of the first and second substances,
Figure BDA0002665003220000023
input variables representing the kth subset of the training set,
Figure BDA0002665003220000024
representing output variables of the kth training set subset, wherein each subset contains I data;
constructing a multi-core parallel random vector function chain network MPVFL model as shown in a formula (1):
Figure BDA0002665003220000025
in the formula (1), θ s Represents the S-th quantile, and S is 1,2, …, S, S is the number of quantiles; z is the number of hidden layer nodes, and J is the number of input nodes; u shape k (θ s ) Denotes the kth subset at the s quantile θ s The following set of weight vectors connecting the input layer and the hidden layer, and having:
Figure BDA0002665003220000026
in the formula (2), the reaction mixture is,
Figure BDA0002665003220000027
indicating that the kth subset is at the jth input level node and the zth hidden level node
Figure BDA0002665003220000028
With weights in between, and having:
Figure BDA0002665003220000029
in the formula (1), V k (θ s ) Denotes the kth subset at the s quantile θ s The following set of connection weight vectors between the hidden layer and the output layer, and has:
Figure BDA0002665003220000031
in the formula (4), the reaction mixture is,
Figure BDA0002665003220000032
indicating that the kth subset is at the z-th hidden layer node
Figure BDA0002665003220000033
And the weight between the output layers;
in the formula (1), W k (θ s ) Denotes the kth subset at the s quantile θ s The set of connection weight vectors between the following input and output layers, and having:
Figure BDA0002665003220000035
in the formula (5), the reaction mixture is,
Figure BDA0002665003220000036
representing the weight of the kth subset between the jth input layer node and the output layer;
in the formula (1), g 1 (. h) represents the activation function of the hidden layer, g 2 () represents the activation function of the output layer;
step 5, K subsets
Figure BDA0002665003220000037
Respectively transmitting the data to K threads, and enabling the K threads to respectively utilize the formula (6) to carry out optimization solution on the formula (1), thereby obtaining theta of K subsets at the s th quantile point s Set of weight vectors { U } between underlying input layer and hidden layer k (θ s ) 1,2, …, K, set of weight vectors { V } between the hidden layer and the output layer k (θ s ) 1,2, …, K and a set of weight vectors { W } between the input layer and the output layer k (θ s ) Parameter estimation value set corresponding to 1,2, …, K |
Figure BDA0002665003220000038
And
Figure BDA0002665003220000039
Figure BDA00026650032200000310
r 'in the formula (7)' 1 、r′ 2 And r' 3 Three penalty parameters of the MPRVFL model;
Figure BDA00026650032200000311
is a loss function and has:
Figure BDA00026650032200000312
in formula (8), μ represents an intermediate variable;
step 6, mixingResults obtained by solving K threads under S quantiles are merged into a parameter set
Figure BDA00026650032200000313
And calculating the parameter set by an asynchronous random gradient descent formula to obtain the optimal weight parameter
Figure BDA00026650032200000314
And
Figure BDA00026650032200000315
step 7, the optimal weight parameter is used
Figure BDA0002665003220000041
And
Figure BDA0002665003220000042
is substituted into formula (1), and the test set (X) is Te ，Y Te ) Input variable X in Te As input of MPRVFL model, obtaining conditional quantile G under S quantiles 1 ，G 2 ，…，G s ，…，G S Wherein G is s Set on the s-th quantile θ for the test s The following predictors, in combination:
Figure BDA0002665003220000043
step 8, predicting values G under all quantites 1 ，G 2 ，…，G S As input variables to the Epanechnikov kernel function; computing the test set (X) using equation (10) Te ，Y Te ) Medium output variable Y Te The result of predicting the runoff probability density of any point q
Figure BDA0002665003220000044
Figure BDA0002665003220000045
In the formula (10), h is the bandwidth, C (. cndot.) is an Epanechnikov kernel function,
step 9, using the test set (X) Te ，Y Te ) All of the output variables Y Te The runoff probability density prediction result is subjected to inverse normalization processing to obtain a runoff prediction value
Figure BDA0002665003220000046
And the probability corresponding to each runoff predicted value
Figure BDA0002665003220000047
Representing the d-th runoff predicted value of the ith time point and the probability value of the d-th runoff predicted value under the ith time point;
step 10, A pieces of
Figure BDA0002665003220000048
Grade of runoff predicted value, and judging the d runoff predicted value of the ith time point
Figure BDA0002665003220000049
Obtaining the levels of the D runoff predicted values at the ith time point, and calculating the sum of the probabilities of all the runoff predicted values in each level as the total probability of the corresponding level; selecting the maximum value of the total probability of each level as the probability of the ith time point, and taking the level of the maximum value of the total probability as the level of the ith time point; further obtaining the levels of 4N +4M-l time points;
step 11, setting two thresholdsnAnd
Figure BDA00026650032200000411
for dividing the A runoff prediction value levels into a water abandoning stage [1,n]load task incomplete stage
Figure BDA00026650032200000410
Counting the number of each stage of the runoff prediction value of 4N +4M-l time pointsAnd measuring to obtain the risk probability value of the occurrence of water abandonment and the incompletion of the load task.
Compared with the prior art, the invention has the beneficial effects that:
1. according to the hydropower station risk assessment method, a large amount of historical actual runoff and related characteristic data are utilized, the uncertainty of the runoff is fully considered, the accuracy and the efficiency of the runoff prediction result are improved, the corresponding risk probability is finally obtained, and the prediction result has reference significance for medium-term and long-term runoff prediction.
2. The RVFL network approach used in the present invention is a variant of an artificial neural network. The method combines the advantages of random weight and function chain, and is a single hidden layer neural network directly connected from an input layer to an output layer. In addition, the direct input and output connection improves the performance of time series prediction, and compared with ARIMA and an artificial neural network, the prediction performance is obviously improved.
3. The MPRVFL model used by the invention adopts data parallel training, averagely divides a training set into a plurality of subsets, then respectively trains the RVFL network model, and obtains the optimal weight parameter by using an asynchronous random gradient descent method. Because the training samples of each prediction model are reduced, the prediction efficiency is improved while the prediction effect is not influenced.
Drawings
FIG. 1 is an overall flow diagram of the present invention;
figure 2 is a schematic diagram of the basic topology of the RVFL network of the present invention.
Detailed Description
In this embodiment, a hydropower station risk assessment method based on multi-core parallel runoff probability density prediction is performed according to the following steps as shown in fig. 1:
step 1, collecting runoff data at different time points and rainfall characteristic, air temperature characteristic and air pressure characteristic related to runoff, and normalizing to obtain a preprocessed runoff data set X ═ X' 1 ，x′ 2 ，…，x′ n ，…，x′ N ) And the precipitation amount feature set P ═ P (P) 1 ，p 2 ，…，p n ，…，p N ) Temperature profile T ═ T 1 ，t 2 ，…，t n ，…，t N ) And set of air pressure characteristics R ═ R (R) 1 ，r 2 ，…，r n ，…，r N ) (ii) a Wherein x is n 、p n 、t n And r n Respectively representing runoff, precipitation, air temperature and air pressure at the ith time point after pretreatment, wherein N is 1,2, …, and N is data quantity acquired by each characteristic;
step 2, predicting the runoff data of the (M +1) th time point by respectively using the runoff data set X' and the data of the first M time points in each characteristic set related to the runoff by using a rolling arrangement prediction method; thereby obtaining an (N-M) X (4M +1) dimensional rolling runoff matrix which is marked as (X, Y); wherein X ═ X (X) 1 ，X 2 ，…，X m ，…，X 4M ) Representing an input variable, X m Represents the m-th input variable, an
Figure BDA0002665003220000051
The jth sample representing the mth input variable, Y ═ Y 1 ，Y 2 ，…，Y m ，…，Y 4N-4M ) T Is an output variable, Y m Represents the mth sample in the output variable Y;
step 3, dividing the first line of the (N-M) X (4M +1) dimension rolling matrix (X, Y) into a training set (X) Tr ，Y Tr ) And the rest as test set (X) Te ，Y Te ) (ii) a Wherein l is more than or equal to 1 and less than 4N-4M;
step 4, training set (X) Tr ，Y Tr ) Divided equally into K subsets
Figure BDA0002665003220000052
Wherein the content of the first and second substances,
Figure BDA0002665003220000053
input variables representing the kth subset of the training set,
Figure BDA0002665003220000061
represents the output variables of the kth subset of the training set, anEach subset contains I data;
combining the basic topology structure diagram of the RVFL network shown in FIG. 2, constructing a multi-core parallel random vector function chain network MPRVFL model shown in formula (1):
Figure BDA0002665003220000062
in the formula (1), θ s Represents the S-th quantile, and S is 1,2, …, S, S is the number of quantiles; z is the number of nodes of the hidden layer, and J is the number of input nodes; u shape k (θ s ) Denotes the kth subset at the s quantile θ s The following set of weight vectors connecting the input layer and the hidden layer, and having:
Figure BDA0002665003220000063
in the formula (2), the reaction mixture is,
Figure BDA0002665003220000064
indicating that the kth subset is at the jth input level node and the zth hidden level node
Figure BDA0002665003220000065
With weights in between, and having:
Figure BDA0002665003220000066
in the formula (1), V k (θ s ) Denotes the kth subset at the s quantile θ s The following set of connection weight vectors between the hidden layer and the output layer, and having:
Figure BDA0002665003220000067
in the formula (4), the reaction mixture is,
Figure BDA0002665003220000068
indicating that the kth subset is at the z-th hidden layer node
Figure BDA0002665003220000069
And the weight between the output layer;
in the formula (1), W k (θ s ) Indicates that the kth subset is at the s quantile point theta s The set of connection weight vectors between the following input and output layers, and having:
Figure BDA00026650032200000611
in the formula (5), the reaction mixture is,
Figure BDA00026650032200000612
representing the weight of the kth subset between the jth input layer node and the output layer;
in the formula (1), g 1 (. h) represents the activation function of the hidden layer, g 2 (-) represents the activation function of the output layer,
step 5, K subsets
Figure BDA00026650032200000613
Respectively transmitting the data to K threads, and enabling the K threads to respectively utilize the formula (6) to carry out optimization solution on the formula (1), thereby obtaining theta of K subsets at the s th quantile point s Set of weight vectors { U } between underlying input layer and hidden layer k (θ s ) 1,2, …, K, set of weight vectors { V } between the hidden layer and the output layer k (θ s ) 1,2, …, K and a set of weight vectors { W } between the input layer and the output layer k (θ s ) Parameter estimation value set corresponding to 1,2, …, K |
Figure BDA0002665003220000071
And
Figure BDA0002665003220000072
Figure BDA0002665003220000073
r 'in the formula (7)' 1 、r′ 2 And r' 3 Three penalty parameters of the MPRVFL model;
Figure BDA0002665003220000074
is a loss function and has:
Figure BDA0002665003220000075
in formula (8), μ represents an intermediate variable;
step 6, merging results obtained by solving K threads under S quantites into a parameter set
Figure BDA0002665003220000076
And calculating a parameter set by an asynchronous random gradient descent formula to obtain an optimal weight parameter
Figure BDA0002665003220000077
And
Figure BDA0002665003220000078
7, optimizing the weight parameters
Figure BDA0002665003220000079
And
Figure BDA00026650032200000710
substituted into formula (1), and test set (X) Te ，Y Te ) Input variable X in Te As the input of MPRVL model, obtaining the conditional quantile G under S quantiles 1 ，G 2 ，…，G s ，…，G S Wherein G is s Set on the s-th quantile θ for the test s The following predictors, in combination:
Figure BDA00026650032200000711
step 8, predicting values G under all quantites 1 ，G 2 ，…，G S As input variables to the Epanechnikov kernel function; computing a test set (X) using equation (10) Te ，Y Te ) Medium output variable Y Te The runoff probability density prediction result of any point q
Figure BDA00026650032200000712
Figure BDA00026650032200000713
In the formula (10), h is the bandwidth, C (·) is the Epanechnikov kernel function,
step 9, use the test set (X) Te ，Y Te ) All of the output variables Y Te The runoff probability density prediction result is subjected to inverse normalization processing to obtain a runoff prediction value
Figure BDA00026650032200000714
And the probability corresponding to each runoff predicted value
Figure BDA0002665003220000081
The d runoff predicted value of the ith time point is shown,
Figure BDA0002665003220000082
representing the probability value of the d runoff predicted value at the ith time point;
step 10, A runoff predicted value grades are set, and each grade is defined as the formula (11):
Figure BDA0002665003220000083
in the formula (11), the reaction mixture is,
Figure BDA0002665003220000084
and
Figure BDA0002665003220000085
the upper limit and the lower limit of runoff of the a-th level are respectively;
judging the d runoff predicted value of the ith time point
Figure BDA0002665003220000086
Obtaining the levels of the D runoff predicted values at the ith time point, and calculating the sum of the probabilities of all the runoff predicted values in each level as the total probability of the corresponding level; selecting the maximum value of the total probability of each level as the probability of the ith time point, and taking the level of the maximum value of the total probability as the level of the ith time point; that is, the probability that the runoff predicted value of the ith time point is in the level of c is
Figure BDA0002665003220000087
Wherein
Figure BDA0002665003220000088
The probability that the runoff predicted value of the ith time point is positioned at the a-th level; further obtaining the levels of 4N +4M-l moment points;
step 11, setting two threshold values according to actual condition regulations of hydropower stationnAnd
Figure BDA0002665003220000089
for dividing the grade of A runoff predicted values into stages of water abandonment [1,n]incomplete stage of load task
Figure BDA00026650032200000810
Counting the grade of the runoff predicted value at all time points and the number n of the runoff predicted values at the water abandoning stage surplus And the number n of incomplete stages of the load task less And dividing the statistical result by the predicted total runoff value quantity N sample Correspondingly obtaining the risk probability of water abandonment and the risk probability of incomplete load tasks; risk rate of water abandonment and incomplete load taskThe risk probability calculation formula is:
Figure BDA00026650032200000811
Figure BDA00026650032200000812
in the formulae (12) and (13),
Figure BDA00026650032200000813
andPrespectively the water abandoning risk rate and the risk probability of incomplete load task, an
Figure BDA00026650032200000814
And N is sample ＝4N+4M-l。

Claims (1)

1. A hydropower station risk assessment method based on multi-core parallel runoff probability density prediction is characterized by comprising the following steps:
step 1, collecting runoff data at different time points and rainfall characteristic, air temperature characteristic and air pressure characteristic related to runoff, and normalizing to obtain a preprocessed runoff data set X ═ X' 1 ,x′ 2 ,…,x′ n ,…,x′ N ) The precipitation feature set P ═ P (P) 1 ,p 2 ,…,p n ,…,p N ) Temperature profile T ═ T 1 ,t 2 ,…,t n ,…,t N ) And set of air pressure characteristics R ═ R (R) 1 ,r 2 ,…,r n ,…,r N ) (ii) a Wherein x is n 、p n 、t n And r n Respectively representing runoff, precipitation, air temperature and air pressure at the ith time point after pretreatment, wherein N is 1,2, …, and N is data quantity acquired by each characteristic;
step 2, predicting the runoff data of the (M +1) th time point by respectively using the runoff data set X' and the data of the first M time points in each characteristic set related to the runoff by using a rolling arrangement prediction method; thereby to obtainObtaining an (N-M) X (4M +1) dimensional rolling runoff matrix, and recording the matrix as (X, Y); wherein X ═ X (X) 1 ,X 2 ,…,X m ,…,X 4M ) Representing an input variable, X m Represents the m-th input variable, an
Figure FDA0002665003210000011
Figure FDA0002665003210000012
The jth sample representing the mth input variable, Y ═ Y 1 ,Y 2 ,…,Y m ,…,Y 4N-4M ) T Is an output variable, Y m Represents the mth sample in the output variable Y;
step 3, dividing the first l rows of the (N-M) X (4M +1) dimension rolling matrix (X, Y) into a training set (X) Tr ,Y Tr ) And the rest as test set (X) Te ,Y Te ) (ii) a Wherein l is more than or equal to 1 and less than 4N-4M;
step 4, training the training set (X) Tr ,Y Tr ) Divided equally into K subsets
Figure FDA0002665003210000013
Wherein the content of the first and second substances,
Figure FDA0002665003210000014
input variables representing the kth subset of the training set,
Figure FDA0002665003210000015
representing output variables of the kth training set subset, wherein each subset contains I data;
constructing a multi-core parallel random vector function chain network MPVFL model as shown in a formula (1):
Figure FDA0002665003210000016
in the formula (1), θ s Denotes the S-th quantile, and S is 1,2, …, S, S is the number of quantiles(ii) a Z is the number of hidden layer nodes, and J is the number of input nodes; u shape k (θ s ) Denotes the kth subset at the s quantile θ s The following set of weight vectors connecting the input layer and the hidden layer, and having:
Figure FDA0002665003210000017
in the formula (2), the reaction mixture is,
Figure FDA0002665003210000018
indicating the kth subset at the jth input layer node and the zth hidden layer node
Figure FDA0002665003210000019
With weights in between, and having:
Figure FDA0002665003210000021
in the formula (1), V k (θ s ) Denotes the kth subset at the s quantile θ s The following set of connection weight vectors between the hidden layer and the output layer, and has:
Figure FDA0002665003210000022
in the formula (4), the reaction mixture is,
Figure FDA0002665003210000023
indicating that the kth subset is at the z-th hidden layer node
Figure FDA0002665003210000024
And the weight between the output layers;
in the formula (1), W k (θ s ) Indicates that the kth subset is at the s quantile point theta s Set of connection weight vectors between input layer and output layer of lower layerAnd has the following components:
Figure FDA0002665003210000025
in the formula (5), the reaction mixture is,
Figure FDA0002665003210000026
representing the weight of the kth subset between the jth input layer node and the output layer;
in the formula (1), g 1 (. h) represents the activation function of the hidden layer, g 2 () represents the activation function of the output layer;
step 5, K subsets
Figure FDA0002665003210000027
Respectively transmitting the data to K threads, and enabling the K threads to respectively utilize the formula (6) to carry out optimization solution on the formula (1), thereby obtaining theta of K subsets at the s th quantile point s Set of weight vectors { U } between underlying input layer and hidden layer k (θ s ) 1,2, …, K, set of weight vectors { V } between the hidden layer and the output layer k (θ s ) 1,2, …, K and a set of weight vectors { W } between the input layer and the output layer k (θ s ) Parameter estimation value set corresponding to 1,2, …, K |
Figure FDA0002665003210000028
And
Figure FDA0002665003210000029
Figure FDA00026650032100000210
r 'in the formula (7)' 1 、r′ 2 And r' 3 Three penalty parameters of the MPRVFL model;
Figure FDA00026650032100000211
is a loss function and has:
Figure FDA00026650032100000212
in formula (8), μ represents an intermediate variable;
step 6, combining results obtained by solving K threads under S quantiles into a parameter set
Figure FDA0002665003210000031
And calculating the parameter set by an asynchronous random gradient descent formula to obtain the optimal weight parameter
Figure FDA0002665003210000032
And
Figure FDA0002665003210000033
step 7, the optimal weight parameter is used
Figure FDA0002665003210000034
And
Figure FDA0002665003210000035
substituted into formula (1), and the test set (X) Te ,Y Te ) Input variable X in Te As the input of MPRVL model, obtaining the conditional quantile G under S quantiles 1 ,G 2 ,…,G s ,…,G S Wherein G is s Set at the s-th quantile θ for the test s The following predictors, in combination:
Figure FDA0002665003210000036
step 8, predicting values G under all quantites 1 ,G 2 ,…,G S As Epanechnikov nucleusAn input variable of the function; computing the test set (X) using equation (10) Te ,Y Te ) Medium output variable Y Te The runoff probability density prediction result of any point q
Figure FDA0002665003210000037
Figure FDA0002665003210000038
In the formula (10), h is the bandwidth, C (. cndot.) is an Epanechnikov kernel function,
step 9, using the test set (X) Te ,Y Te ) All of the output variables Y Te The runoff probability density prediction result is subjected to inverse normalization processing to obtain a runoff prediction value
Figure FDA0002665003210000039
And the probability corresponding to each runoff predicted value
Figure FDA00026650032100000310
Figure FDA00026650032100000311
The d-th runoff predicted value of the i-th time point is shown,
Figure FDA00026650032100000312
representing the probability value of the d runoff predicted value at the ith time point;
step 10, setting A runoff predicted value grades, and judging the d runoff predicted value of the i-th time point
Figure FDA00026650032100000313
Obtaining the levels of the D runoff predicted values at the ith time point, and calculating the sum of the probabilities of all the runoff predicted values in each level as the total probability of the corresponding level; selecting the maximum value of the total probability of each level as the ith timeThe probability of the moment point, and the level of the maximum value of the total probability is taken as the level of the ith moment point; further obtaining the levels of 4N +4M-l time points;
step 11, setting two thresholdsnAnd
Figure FDA00026650032100000314
for dividing the A runoff predictor levels into a water abandonment stage [1,n]incomplete stage of load task
Figure FDA00026650032100000315
And counting the number of each stage where the runoff predicted values of 4N +4M-l time points are located, thereby obtaining the risk probability value of the occurrence of water abandonment and the incompletion of the load task.
CN202010915902.3A 2020-09-03 2020-09-03 Hydropower station risk assessment method based on multi-core parallel runoff probability density prediction Active CN112036649B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010915902.3A CN112036649B (en) 2020-09-03 2020-09-03 Hydropower station risk assessment method based on multi-core parallel runoff probability density prediction

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010915902.3A CN112036649B (en) 2020-09-03 2020-09-03 Hydropower station risk assessment method based on multi-core parallel runoff probability density prediction

Publications (2)

Publication Number Publication Date
CN112036649A CN112036649A (en) 2020-12-04
CN112036649B true CN112036649B (en) 2022-09-13

Family

ID=73592337

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010915902.3A Active CN112036649B (en) 2020-09-03 2020-09-03 Hydropower station risk assessment method based on multi-core parallel runoff probability density prediction

Country Status (1)

Country Link
CN (1) CN112036649B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104182807A (en) * 2014-08-21 2014-12-03 大连理工大学 Reservoir dispatching risk evaluation method by considering runoff forecast uncertainty
WO2015172560A1 (en) * 2014-05-16 2015-11-19 华南理工大学 Central air conditioner cooling load prediction method based on bp neural network
CN105243502A (en) * 2015-10-19 2016-01-13 华中科技大学 Hydropower station scheduling risk assessment method and system based on runoff interval prediction
CN106295899A (en) * 2016-08-17 2017-01-04 合肥工业大学 Based on genetic algorithm and the wind power probability density Forecasting Methodology supporting vector quantile estimate
CN110458722A (en) * 2019-07-25 2019-11-15 淮阴工学院 Flood interval prediction method based on multiple target random vector function connection network

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2015172560A1 (en) * 2014-05-16 2015-11-19 华南理工大学 Central air conditioner cooling load prediction method based on bp neural network
CN104182807A (en) * 2014-08-21 2014-12-03 大连理工大学 Reservoir dispatching risk evaluation method by considering runoff forecast uncertainty
CN105243502A (en) * 2015-10-19 2016-01-13 华中科技大学 Hydropower station scheduling risk assessment method and system based on runoff interval prediction
CN106295899A (en) * 2016-08-17 2017-01-04 合肥工业大学 Based on genetic algorithm and the wind power probability density Forecasting Methodology supporting vector quantile estimate
CN110458722A (en) * 2019-07-25 2019-11-15 淮阴工学院 Flood interval prediction method based on multiple target random vector function connection network

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
A novel wind power probabilistic forecasting approach based on joint quantile regression and multi-objective optimization;Hu, Jianming .et;《RENEWABLE ENERGY》;20200421;全文 *
Probability density forecasting of wind power using quantile regression neural network and kernel density estimation;He, Yaoyao .et;《ENERGY CONVERSION AND MANAGEMENT》;20180515;全文 *
基于Epanechnikov核与最优窗宽组合的中期电力负荷概率密度预测方法;何耀耀等;《电力自动化设备》;20161110(第11期);全文 *
基于分位数回归的降水、径流变化及响应分析;商颂;《中国优秀硕士学位论文全文数据库 (工程科技Ⅱ辑)》;20171115;全文 *
改进鲸鱼算法优化混合核支持向量机在径流预测中的应用;周有荣等;《中国农村水利水电》;20200715(第07期);全文 *

Also Published As

Publication number Publication date
CN112036649A (en) 2020-12-04

Similar Documents

Publication Publication Date Title
CN112949945B (en) Wind power ultra-short-term prediction method for improving bidirectional long-term and short-term memory network
CN102478584B (en) Wind power station wind speed prediction method based on wavelet analysis and system thereof
CN108009667A (en) A kind of energy demand total amount and structure prediction system
CN113128113B (en) Lean information building load prediction method based on deep learning and transfer learning
CN102479347B (en) Method and system for forecasting short-term wind speed of wind farm based on data driving
CN111027775A (en) Step hydropower station generating capacity prediction method based on long-term and short-term memory network
CN104408562A (en) Photovoltaic system generating efficiency comprehensive evaluation method based on BP (back propagation) neural network
CN103971175B (en) Short-term load prediction method of multistage substations
CN111353652A (en) Wind power output short-term interval prediction method
CN112100911B (en) Solar radiation prediction method based on depth BILSTM
CN105809349A (en) Scheduling method considering incoming water correlation cascade hydropower stations
CN111695666A (en) Wind power ultra-short term conditional probability prediction method based on deep learning
Kofinas et al. Daily multivariate forecasting of water demand in a touristic island with the use of artificial neural network and adaptive neuro-fuzzy inference system
CN115034485A (en) Wind power interval prediction method and device based on data space
CN103617447A (en) Evaluation system and method for intelligent substation
CN115758151A (en) Combined diagnosis model establishing method and photovoltaic module fault diagnosis method
CN115859099A (en) Sample generation method and device, electronic equipment and storage medium
CN110555566B (en) B-spline quantile regression-based photoelectric probability density prediction method
CN112036649B (en) Hydropower station risk assessment method based on multi-core parallel runoff probability density prediction
CN111798055A (en) Variable weight combined photovoltaic output prediction method based on grey correlation degree
CN110414734A (en) A method of meter and the assessment of wind-resources usage forecast
CN116341705A (en) Long-period memory network water quality parameter prediction method based on sparse label
Yang Short-term load monitoring of a power system based on neural network
CN114037272A (en) Energy efficiency assessment method for regional comprehensive energy system
CN110070209B (en) Short-term load prediction method of district heating system based on SD-DNNs

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