CN104090974B - The Application of Data Mining of the follow-up water of extension reservoir and system - Google Patents

Abstract

The Application of Data Mining of the present invention a kind of extension follow-up water of reservoir and system, described method comprises the steps: by total Water size in schedule periods, historical flood is carried out row's frequency, according to total Water frequency threshold, historical flood is divided into several magnitudes；Historical flood data is made standardization；Select flood similarity indices, and each index is combined, form different flood similarity indices assembled schemes；Similar flood Dynamic Recognition, builds similar flood collection；Similar flood is chosen；Similar flood extension.The present invention has estimated follow-up water process and the water yield of reservoir by the Dynamic Recognition of similar flood and extension, effectively compensate for the problem that current flood forecasting leading time is too short.

Description

The Application of Data Mining of the follow-up water of extension reservoir and system

Technical field

The invention belongs to the reservoir Technique for Real-time Joint Operation of Flood technology in hydraulic engineering field, after particularly relating to a kind of extension reservoir The Application of Data Mining of continuous water and system.

Background technology

For Technique for Real-time Joint Operation of Flood, the prediction of the follow-up water of reservoir is significant to it, it is therefore desirable to combine The achievement of Real-time Flood Forecasting instructs Flood Control Dispatch decision-making.But it being limited to technical merit, current flood forecasting can only have Be given in limit leading time and meet the flood forecasting of precision as a result, it is difficult to meet the requirement of reservoir Technique for Real-time Joint Operation of Flood.

The problem too short in order to solve flood forecasting leading time, currently mainly has two approach:

Article 1, approach: the method using short-term weather prediction and product Confluence Model to be coupled carries out flood forecasting, the party Although method can extend the leading time of flood forecasting to a certain extent, it is contemplated that the space-time distribution of rainfall has stronger Randomness feature, it applies less effective.

Article 2 approach: sorting out historical flood and flood similarity identification, it is complete that existing research is based on flood Process, selected similarity evaluation index (heavy rain shrouds area, heavy rain always lasts, return period of flood, flood composition etc.) is all It is could to obtain after flood overall process, it is impossible to meet the requirement of reservoir Technique for Real-time Joint Operation of Flood, not there is practical value.

Additionally, different magnitude of flood exists different inherent mechanism difference, it is embodied in different magnitude of flood and exists There is in the factors such as prophase programming spatial and temporal distributions, interval inflow, antecedent soil moisture bigger difference, and existing technology is equal It is not directed to the similar flood Study of recognition of component level.At part Lack-data watershed, owing to the data of flood similarity indices is the completeest Standby, therefore identification and the extension of the similar flood of Lack-data watershed is also the problem needing solution badly.

Summary of the invention

Goal of the invention a: purpose is to provide the Application of Data Mining of a kind of follow-up water of extension reservoir, to solve The problem that current flood forecasting leading time is too short and cannot meet the requirement of reservoir Technique for Real-time Joint Operation of Flood.

Further objective is that the Dynamic Data Mining system that a kind of follow-up water of extension reservoir is provided, to realize above-mentioned side Method.

Technical scheme: the Application of Data Mining of a kind of follow-up water of extension reservoir, comprises the steps:

Historical flood is carried out row's frequency by total Water size in schedule periods, according to total Water frequency threshold, historical flood is divided For several magnitudes；Historical flood data is made standardization；Select flood similarity indices, and each index is carried out group Close, form different flood similarity indices assembled schemes；Similar flood Dynamic Recognition, builds similar flood collection；Similar flood Choose；Similar flood extension.

In further embodiment, comprise the steps:

Step 1: historical flood is carried out row's frequency by total Water size in schedule periods, according to total Water frequency threshold by history Flood is divided into several magnitudes, by occurred the period accumulation water yield of duration in historical flood data base equal duration time In section accumulation water yield sequence, row's frequency, carries out similar flood retrieval in corresponding data, obtains relevant historical flood data；

Step 2: historical flood data is done standardization, using each historical flood rainfall start time as former Point, is converted to sequential by the absolute time that historical flood occurs；

Step 3: select flood similarity indices, and each index is combined, form different flood similaritys and refer to Mark assembled scheme, described flood similar figures index has included rise flow, period accumulation rainfall and the period accumulation water yield；

Step 4: similar flood Dynamic Recognition, builds similar flood collection, and some for history floods are each considered as a class, meter Calculate all kinds of between distance, two minimum for distance classes will be merged into a new class, then under the differentiation of new classification, again Calculate all kinds of between distance, then two minimum for distance classes will be merged into a new class, until all patterns are polymerized to two classes and are Only；

Step 5: choose similar flood；Rise by flood section if being in or still have bigger follow-up rainfall, then concentrating from similar flood Select last similar flood long, that magnanimity is big or take the outer envelope curve of similar flood collection；If being in the afterbody of rainfall, then from phase Concentrate like flood and select last similar flood short, that magnanimity is little or take the interior envelope curve of similar flood collection；Other situations then select phase Averaging process like flood collection；

Step 6: similar flood extension；In Real-time Flood Forecasting leading time, use the achievement of Real-time Flood Forecasting；In advance See outside the phase, concentrate from similar flood and select similar flood to splice.

The step building similar flood collection farther includes:

Step 41, initialization classification, choose the flood that history N field data is the most complete, calculate each field K similarity of flood Index value, is each considered as a class by history N field flood, is designated as: G_i ⁽⁰⁾={ x_iI=1,2 ..., N, N, K are natural number；

Step 42, calculate all kinds of between Euclidean distance D_{I, j}, generate a full symmetric Distance matrix D^(K)= (D_{I, j})_m×m, wherein, m is the number of class, m=N time initial；Euclidean distance D_{I, j}Below equation can be used to calculate:

$D_{i,j}=\sqrt{{({Q_0}^i-{Q_0}^j)}^2+{({P_t}^i-{P_t}^j)}^2+{({W_t}^i-{W_t}^j)}^2}$

Matrix D in step 43, calculation procedure 42^(K)In least member, if it is G_a ^(K)And G_b ^(K)Between distance, by G_a ^(K) And G_b ^(K)Being merged into a class, the newest cluster is designated as: G₁ ^(K+1), G₂ ^(K+1)..., make k=k+1, m=m-1；

Step 44, the number of inspection class, such as number m of fruit > 2, go to step 42 and proceed cluster calculation；Otherwise, Calculating terminates；

Step 45, build similar flood collection, according to the result of Hierarchical clustering analysis, take front some floods and build similar flood Water collection.

Described similar flood extension step farther includes:

When flood splicing moment real-time prediction value is inconsistent to similar flood discharge value, use integral translation phase Like the method realization smooth connection of peb process, below equation can be used to calculate:

$Q_2\left(t\right)=\left\{\begin{array}{cc}Q\left(t\right)& t\≤t_0+\mathrm{\τ}\\ Q_1\left(t\right)+\mathrm{\ΔQ}& t>t_0+\mathrm{\τ}\end{array}\right.$

Δ Q=Q (t₀+τ)-Q₁(t₀+τ)

Q₀The flow that rises is acted for flood；

t₀For current time；

τ is Real-time Flood Forecasting leading time；

Q (t) is the discharge process of Real-time Flood Forecasting；

Q₁T () is the similar flood discharge process chosen；

Q₂T () is similar flood extension discharge process；

Δ Q is t₀The difference of+τ moment real-time prediction flow value and similar flood discharge value.

In a further embodiment, step 7 is also included:

Evaluate flood extension effect；Evaluate flood extension by follow-up water yield extension ratio of profit increase and deterministic coefficient increment to imitate Really；

Follow-up water yield extension ratio of profit increase uses below equation to calculate:

Wherein, Q_RealT () is observed flood process；T is schedule periods, t₀For current time, Q (t) is Real-time Flood Forecasting Discharge process；Q₂T () is similar flood extension discharge process；

Deterministic coefficient increment uses below equation to calculate:

Δ DC=DC₂-DC

Wherein, DC₂Deterministic coefficient for similarity extension peb process；DC is the definitiveness of real-time prediction peb process Coefficient；

It is respectively adopted below equation to calculate:

Wherein,For the meansigma methods of observed flood process, T is schedule periods, t₀For current time.

A kind of Dynamic Data Mining system of the follow-up water of extension reservoir, including such as lower module:

Row's frequency division generic module, for carrying out row's frequency to historical flood by total Water size in schedule periods, and according to total Water Historical flood is divided into several magnitudes by frequency threshold；

Standardization module, for making standardization to historical flood data；

Similarity indices chooses module, is used for selecting flood similarity indices, and each index is combined, and is formed not Same flood similarity indices assembled scheme；

Similar flood Dynamic Recognition module, for the Dynamic Recognition of similar flood,

Similar flood collection builds module, is used for building similar flood collection；

Similar flood chooses module, for choosing of similar flood；

Similar flood extension module, for similar flood extension.

Further technical scheme is:

Row's frequency division generic module: for historical flood is carried out row's frequency by total Water size in schedule periods, according to total Water frequency Historical flood is divided into several magnitudes by rate threshold value, the period accumulation water yield of duration will have occurred in historical flood data base In the period accumulation water yield sequence of equal duration, row's frequency, carries out similar flood retrieval in corresponding data, obtains relevant historical Flood data；

Standardization module: for historical flood data is done standardization, open with each historical flood rainfall The absolute time that historical flood occurs, as initial point, is converted to sequential by moment beginning；

Similarity indices chooses module: is used for selecting flood similarity indices, and each index is combined, and is formed not With flood similarity indices assembled scheme, described flood similar figures index included rising flow, period accumulation rainfall and time The section accumulation water yield；

Similar flood Dynamic Recognition module: for similar flood Dynamic Recognition, builds similar flood collection, by some for history fields Flood is each considered as a class, calculate all kinds of between distance, two minimum for distance classes will be merged into a new class, then newly Classification differentiation under, recalculate all kinds of between distance, then two minimum for distance classes will be merged into a new class, until Till all patterns are polymerized to two classes；

Similar flood chooses module: be used for choosing similar flood, specifically, if being in the flood section or still have bigger follow-up of rising Rainfall, then concentrate from similar flood and select last similar flood long, that magnanimity is big or take the outer envelope curve of similar flood collection；If being in The afterbody of rainfall, then concentrate from similar flood and select last similar flood short, that magnanimity is little or take in similar flood collection Envelope curve；Other situations then select the averaging process of similar flood collection；

Similar flood extension module: extend for similar flood, in Real-time Flood Forecasting leading time, uses real-time flood The achievement of forecast；Outside leading time, concentrate from similar flood and select similar flood to splice.

Build similar flood collection module to farther include:

Step 41, initialization classification, choose the flood that history N field data is the most complete, calculate each field K similarity of flood Index value.History N field flood is each considered as a class, is designated as: G_i ⁽⁰⁾={ x_iI=1,2 ..., N, N, K are natural number；

Step 42, calculate all kinds of between Euclidean distance D_{I, j}, generate a full symmetric Distance matrix D^(K)= (D_{I, j})_m×m, wherein, m is the number of class, m=N time initial；Euclidean distance D_{I, j}Below equation can be used to calculate:

$D_{i,j}=\sqrt{{({Q_0}^i-{Q_0}^j)}^2+{({P_t}^i-{P_t}^j)}^2+{({W_t}^i-{W_t}^j)}^2}$

Matrix D in step 43, calculation procedure 42^(K)In least member, if it is G_a ^(K)And G_b ^(K)Between distance, by G_a ^(K) And G_b ^(K)Being merged into a class, the newest cluster is designated as: G₁ ^(K+1), G₂ ^(K+1)..., make k=k+1, m=m-1；

Step 44, the number of inspection class, such as number m of fruit > 2, go to step 42 and proceed cluster calculation；Otherwise, Calculating terminates；

Step 45, build similar flood collection, according to the result of Hierarchical clustering analysis, take front some floods and build similar flood Water collection.

Described similar flood extension module farther includes:

When flood splicing moment real-time prediction value is inconsistent to similar flood discharge value, use integral translation phase Like the method realization smooth connection of peb process, below equation can be used to calculate:

$Q_2\left(t\right)=\left\{\begin{array}{cc}Q\left(t\right)& t\≤t_0+\mathrm{\τ}\\ Q_1\left(t\right)+\mathrm{\ΔQ}& t>t_0+\mathrm{\τ}\end{array}\right.$

Δ Q=Q (t₀+τ)-Q₁(t₀+τ)

Q₀The flow that rises is acted for flood；

t₀For current time；

τ is Real-time Flood Forecasting leading time；

Q (t) is the discharge process of Real-time Flood Forecasting；

Q₁T () is the similar flood discharge process chosen；

Q₂T () is similar flood extension discharge process；

Δ Q is t₀The difference of+τ moment real-time prediction flow value and similar flood discharge value.

In further technical scheme, also include flood extension effect assessment module, for by follow-up water yield extension Ratio of profit increase and deterministic coefficient increment evaluate flood extension effect；

Wherein, follow-up water yield extension ratio of profit increase uses below equation to calculate:

Wherein, Q_RealT () is observed flood process；T is schedule periods, t₀For current time, Q (t) is Real-time Flood Forecasting Discharge process；Q₂T () is similar flood extension discharge process；

Deterministic coefficient increment uses below equation to calculate:

Δ DC=DC₂-DC

Wherein, DC₂Deterministic coefficient for similarity extension peb process；DC is the definitiveness of real-time prediction peb process Coefficient；

It is respectively adopted below equation to calculate:

Wherein,For the meansigma methods of observed flood process, T is schedule periods, t₀For current time.

Beneficial effect: the present invention by the Dynamic Recognition of similar flood and extension estimated reservoir follow-up water process and The water yield, compensate for the problem that current flood forecasting leading time is too short effectively.Further in technical scheme, the present invention considers The characteristic that storm flood information dynamically increases, from the storm flood information progressively presented, develops to the flood that reservoir is follow-up Trend is estimated, it is possible to preferably meet the requirement of reservoir Technique for Real-time Joint Operation of Flood, has important practical value.It addition, this Historical flood is divided into several magnitudes by invention, it is contemplated that different magnitude of flood exists the difference of different inherent mechanism, it is considered to Factor is than prior art more fully.Finally, the present invention is preferred by the combination of flood similarity indices, simplifies flood similar Property index system, similar flood identification and extension for Lack-data watershed provide a new approach, decrease meter simultaneously Operator workload.

Accompanying drawing explanation

Fig. 1 is the flow chart of the inventive method.

Fig. 2 is the flood clustering figure of hierarchical clustering method of the present invention.

Fig. 3 is that the similar flood of the present invention splices schematic diagram.

Wherein: Q₀The flow that rises is acted for flood；t₀For current time；τ is Real-time Flood Forecasting leading time；Q (t) is the most big vast The discharge process of water forecast；Q₁T () is the similar flood discharge process chosen；Q₂T () is similar flood extension discharge process；ΔQ For t₀The difference of+τ moment real-time prediction flow value and similar flood discharge value.

Fig. 4 is similar flood extension schematic diagram.Wherein: T is schedule periods.

Detailed description of the invention

In conjunction with Fig. 1 to Fig. 4, the present invention is described.Historical flood is divided into multiple magnitude by the present invention, the most large, medium and small three Magnitude, has formulated the standardized principle of historical flood, constructs flood similarity indices system, carries out based on hierarchical clustering principle Similar flood Dynamic Recognition and extension, by evaluating flood extension effect, the assembled scheme to different flood similarity indices Carry out preferably.

The Application of Data Mining of the follow-up water of extension reservoir of the present invention, comprises the following steps:

Step 1, to historical flood by total Water size in schedule periods carry out row frequency, according to total Water frequency threshold (such as: 85%, 40%) historical flood is divided into large, medium and small three magnitudes, the period accumulation water yield of duration will occur at history flood Row's frequency (from big to small) in the period accumulation water yield sequence of equal duration in water data base, corresponding frequency is designated as P_t.If P_t＜ 40%, then in big magnitude historical flood data base, carry out similar flood retrieval；If 40%≤P_t＜ 85%, then go through at middleweight History flood data base carries out similar flood retrieval；If P_t>=85%, then carry out similar in little magnitude historical flood data base Flood is retrieved.

Step 2, makees standardization to historical flood data:

In view of history difference play flood time of origin, always lasting and the diversity of flood form, the inventive method is adopted With unified relative time coordinate, using each historical flood rainfall start time as initial point, and historical flood is occurred Absolute time be converted to sequential.

Step 3, selects flood similarity indices, and each index is combined, form different flood similaritys and refer to Mark assembled scheme:

Confluxing theory according to product, the principal element affecting peb process includes rainfall and space-time distribution, early stage soil Water content and land surface condition, the inventive method combines real data situation and selects three below index structure flood similarity to refer to Mark system:

(1) the flow Q that rises is acted₀.Antecedent soil moisture is the important parameter of flow anomaly, but antecedent soil moisture Monitoring materials extremely lacks, and what the inventive method was chosen act the flow indicator that rises can characterize early stage soil to a certain extent indirectly The size of water content.

(2) period accumulation rainfall P_t.Rainfall and space-time distribution thereof determine flood size and shape to a certain extent State, period accumulation rainfall can reflect the size of rainfall and the variation characteristic of rainfall simultaneously, can use below equation meter Calculate:

$P_t=\underset{i=1}{\overset{t}{\mathrm{\Σ}}}{P}_i---\left(1\right)$

Wherein, t is the sequential of current time；P_iFor hourly precipitation amount；Other variable is the same.

(3) period accumulation water yield W_t.The watershed system conditional decision difference of rainfall and peb process, and water Storehouse Technique for Real-time Joint Operation of Flood more pays close attention to the size of magnanimity, and therefore the inventive method selects the period accumulation water yield to reflect flood occurred The form of water and the land surface condition in basin, can use below equation to calculate:

$W_t=\underset{i=1}{\overset{t}{\mathrm{\Σ}}}{Q}_i\·\mathrm{\Δt}---\left(2\right)$

Wherein, Q_iFor period reservoir inflow；Segment length when Δ t is；Other variable is the same.

Situation about cannot obtain for the data of Lack-data watershed Partial Flood similarity indices, the present invention is by above three Flood similarity indices is combined, and constitutes Q altogether₀、P_t、W_t、(Q₀, P_t)、(Q₀, W_t)、(P_t, W_t)、(Q₀, P_t, W_t) 7 different Flood similarity indices assembled scheme, carries out excellent by extension effect assessment to the assembled scheme of different flood similarity indices Choosing.For Lack-data watershed, can select and full indicator combination (Q₀, P_t, W_t) simplification indicator combination that extension effect is close (depending on Concrete data situation with depending on extension effect) carry out flood similar flood identification and extension.

Step 4, similar flood Dynamic Recognition, build similar flood collection:

The inventive method carries out the Dynamic Recognition of similar flood based on hierarchical clustering principle, first by each for history N field flood From being considered as a class, calculate all kinds of between distance, two minimum for distance classes will be merged into a new class, then in new classification Differentiation under, recalculate all kinds of between distance, then two minimum for distance classes will be merged into a new class, until all moulds Till formula is polymerized to two classes, specifically include following five sub-steps:

(1) classification is initialized.Choose the flood that history N field data is the most complete, calculate each field K similarity indices of flood Numerical value.History N field flood is each considered as a class, is designated as: G_i ⁽⁰⁾={ x_iI=1,2 ..., N.

(2) calculate all kinds of between Euclidean distance D_{I, j}, generate a full symmetric Distance matrix D^(K)=(D_{I, j})_m×m, Wherein, m is the number (m=N time initial) of class.Wherein, Euclidean distance D_{I, j}Below equation can be used to calculate:

$D_{i,j}=\sqrt{{({Q_0}^i-{Q_0}^j)}^2+{({P_t}^i-{P_t}^j)}^2+{({W_t}^i-{W_t}^j)}^2}---\left(3\right)$

(3) matrix D in calculation procedure (2)^(K)In least member, if it is G_a ^(K)And G_b ^(K)Between distance, by G_a ^(K)With G_b ^(K)Being merged into a class, the newest cluster is designated as: G₁ ^(K+1),G₂ ^(K+1)..., make k=k+1, m=m-1.

(4) number of class is checked.Number m such as fruit > 2, go to step (2) and proceed cluster calculation；Otherwise, calculate Terminate.

(5) similar flood collection is built.It is illustrated in figure 2 the flood clustering figure of hierarchical clustering method, according to Hierarchical clustering analysis Result, take front n field flood and build similar flood collection.

Step 5, similar flood chooses:

Choosing mainly with the current flood-proofing terrain of reservoir as foundation of similar flood, if being in the flood section or still have bigger of rising Follow-up rainfall, then concentrate from similar flood and select last similar flood long, that magnanimity is big or take the outer envelope curve of similar flood collection；If It is in the afterbody of rainfall, then concentrates from similar flood and select last similar flood short, that magnanimity is little or take similar flood collection Interior envelope curve；The averaging process of similar flood collection then can be selected for other situations.

Step 6, similar flood extension:

It is illustrated in figure 3 similar flood splicing schematic diagram, wherein: Q₀The flow that rises is acted for flood；t₀For current time；τ is real Time flood forecasting leading time；Q (t) is the discharge process of Real-time Flood Forecasting；Q₁T () is the similar flood discharge process chosen；Q₂ T () is similar flood extension discharge process；Δ Q is t₀The difference of+τ moment real-time prediction flow value and similar flood discharge value.

The principle of similar flood extension is: in Real-time Flood Forecasting leading time τ, uses the achievement of Real-time Flood Forecasting； Outside leading time τ, concentrate from similar flood and select similar flood to splice.For flood splicing moment real-time prediction value and phase Like the flood discharge inconsistent situation of value, use the method for the similar peb process of integral translation to realize linking up smoothly, can use with Lower formula calculates:

$Q_2\left(t\right)=\left\{\begin{array}{cc}Q\left(t\right)& t\≤t_0+\mathrm{\τ}\\ Q_1\left(t\right)+\mathrm{\ΔQ}& t>t_0+\mathrm{\τ}\end{array}\right.---\left(4\right)$

Δ Q=Q (t₀+τ)-Q₁(t₀+τ) (5)

Step 7, flood extension effect assessment:

Can be obtained the reservoir follow-up water process of certain precision by above step, flood extension is imitated by the inventive method The evaluation of fruit mainly selects following two evaluation index:

(1) follow-up water yield extension ratio of profit increase λ.The size of the follow-up water yield be reservoir Technique for Real-time Joint Operation of Flood paid close attention to important because of Element, the reflection of follow-up water yield extension ratio of profit increase is the lifting journey that the follow-up water yield is estimated after flood similarity extension ability Degree, can use below equation to calculate:

Wherein, Q_RealT () is observed flood process；T is schedule periods.

(2) deterministic coefficient increment Delta DC.Deterministic coefficient reflects the fitting degree between peb process, definitiveness system Number increment reflection is the raising value of deterministic coefficient after flood similarity extension, and below equation can be used to count Calculate:

Δ DC=DC₂-DC (7)

Wherein, DC₂Deterministic coefficient for similarity extension peb process；DC is the definitiveness of real-time prediction peb process Coefficient, is respectively adopted below equation and calculates:

Wherein,Meansigma methods for observed flood process.

Follow-up water yield extension ratio of profit increase and deterministic coefficient increment are the biggest more excellent type evaluation index, and index value of calculation is more Greatly, show that the effect of similar flood extension is the best, the biggest to the meaning of reservoir Technique for Real-time Joint Operation of Flood.

In order to realize said method, build following system:

A kind of Dynamic Data Mining system of the follow-up water of extension reservoir, including such as lower module:

Row's frequency division generic module, for carrying out row's frequency to historical flood by total Water size in schedule periods, and according to total Water Historical flood is divided into several magnitudes by frequency threshold；

Standardization module, for making standardization to historical flood data；

Similarity indices chooses module, is used for selecting flood similarity indices, and each index is combined, and is formed not Same flood similarity indices assembled scheme；

Similar flood Dynamic Recognition module, for the Dynamic Recognition of similar flood,

Similar flood collection builds module, is used for building similar flood collection；

Similar flood chooses module, for choosing of similar flood；

Similar flood extension module, for similar flood extension.

Further technical scheme is:

Row's frequency division generic module: for historical flood is carried out row's frequency by total Water size in schedule periods, according to total Water frequency Historical flood is divided into several magnitudes by rate threshold value, the period accumulation water yield of duration will have occurred in historical flood data base In the period accumulation water yield sequence of equal duration, row's frequency, carries out similar flood retrieval in corresponding data, obtains relevant historical Flood data；

Standardization module: for historical flood data is done standardization, open with each historical flood rainfall The absolute time that historical flood occurs, as initial point, is converted to sequential by moment beginning；

Similarity indices chooses module: is used for selecting flood similarity indices, and each index is combined, and is formed not With flood similarity indices assembled scheme, described flood similar figures index included rising flow, period accumulation rainfall and time The section accumulation water yield；

Similar flood Dynamic Recognition module: for similar flood Dynamic Recognition, builds similar flood collection, by some for history fields Flood is each considered as a class, calculate all kinds of between distance, two minimum for distance classes will be merged into a new class, then newly Classification differentiation under, recalculate all kinds of between distance, then two minimum for distance classes will be merged into a new class, until Till all patterns are polymerized to two classes；

Similar flood chooses module: be used for choosing similar flood, specifically, if being in the flood section or still have bigger follow-up of rising Rainfall, then concentrate from similar flood and select last similar flood long, that magnanimity is big or take the outer envelope curve of similar flood collection；If being in The afterbody of rainfall, then concentrate from similar flood and select last similar flood short, that magnanimity is little or take in similar flood collection Envelope curve；Other situations then select the averaging process of similar flood collection；

Similar flood extension module: extend for similar flood, in Real-time Flood Forecasting leading time, uses real-time flood The achievement of forecast；Outside leading time, concentrate from similar flood and select similar flood to splice.

Build similar flood collection module to farther include:

Step 41, initialization classification, choose the flood that history N field data is the most complete, calculate each field K similarity of flood Index value.History N field flood is each considered as a class, is designated as: G_i ⁽⁰⁾={ x_iI=1,2 ..., N, N, K are natural number；

Step 42, calculate all kinds of between Euclidean distance D_{I, j}, generate a full symmetric Distance matrix D^(K)= (D_{I, j})_m×m, wherein, m is the number of class, m=N time initial；Euclidean distance D_{I, j}Below equation can be used to calculate:

$D_{i,j}=\sqrt{{({Q_0}^i-{Q_0}^j)}^2+{({P_t}^i-{P_t}^j)}^2+{({W_t}^i-{W_t}^j)}^2}---\left(3\right)$

Matrix D in step 43, calculation procedure 42^(K)In least member, if it is G_a ^(K)And G_b ^(K)Between distance, by G_a ^(K) And G_b ^(K)Being merged into a class, the newest cluster is designated as: G₁ ^(K+1),G₂ ^(K+1)..., make k=k+1, m=m-1；

Step 44, the number of inspection class, such as number m of fruit > 2, go to step 42 and proceed cluster calculation；Otherwise, Calculating terminates；

Step 45, build similar flood collection, according to the result of Hierarchical clustering analysis, take front some floods and build similar flood Water collection.

Described similar flood extension module farther includes:

When flood splicing moment real-time prediction value is inconsistent to similar flood discharge value, use integral translation phase Like the method realization smooth connection of peb process, below equation can be used to calculate:

$Q_2\left(t\right)=\left\{\begin{array}{cc}Q\left(t\right)& t\≤t_0+\mathrm{\τ}\\ Q_1\left(t\right)+\mathrm{\ΔQ}& t>t_0+\mathrm{\τ}\end{array}\right.$

Δ Q=Q (t₀+τ)-Q₁(t₀+τ)

Q₀The flow that rises is acted for flood；

t₀For current time；

τ is Real-time Flood Forecasting leading time；

Q (t) is the discharge process of Real-time Flood Forecasting；

Q₁T () is the similar flood discharge process chosen；

Q₂T () is similar flood extension discharge process；

Δ Q is t₀The difference of+τ moment real-time prediction flow value and similar flood discharge value.

In further technical scheme, also include flood extension effect assessment module, for by follow-up water yield extension Ratio of profit increase and deterministic coefficient increment evaluate flood extension effect；

Wherein, follow-up water yield extension ratio of profit increase uses below equation to calculate:

Wherein, Q_RealT () is observed flood process；T is schedule periods, t₀For current time, Q (t) is Real-time Flood Forecasting Discharge process；Q₂T () is similar flood extension discharge process；

Deterministic coefficient increment uses below equation to calculate:

Δ DC=DC₂-DC

Wherein, DC₂Deterministic coefficient for similarity extension peb process；DC is the definitiveness of real-time prediction peb process Coefficient；

It is respectively adopted below equation to calculate:

Wherein,For the meansigma methods of observed flood process, T is schedule periods, t₀For current time.

The preferred embodiment of the present invention described in detail above, but, the present invention is not limited in above-mentioned embodiment Detail, in the technology concept of the present invention, technical scheme can be carried out multiple equivalents, this A little equivalents belong to protection scope of the present invention.

It is further to note that each the concrete technical characteristic described in above-mentioned detailed description of the invention, at not lance In the case of shield, can be combined by any suitable means.In order to avoid unnecessary repetition, the present invention to various can The compound mode of energy illustrates the most separately.

Additionally, combination in any can also be carried out between the various different embodiment of the present invention, as long as it is without prejudice to this The thought of invention, it should be considered as content disclosed in this invention equally.

Claims (2)

1. the Application of Data Mining of the follow-up water of extension reservoir, it is characterised in that comprise the steps:

Step 1: historical flood is carried out row's frequency by total Water size in schedule periods, according to total Water frequency threshold by historical flood It is divided into several magnitudes, the period accumulation water yield period of equal duration in historical flood data base duration occurred is tired out Row's frequency in water accumulating volume sequence, carries out similar flood retrieval in corresponding data, obtains relevant historical flood data；

Step 2: historical flood data is done standardization, using each historical flood rainfall start time as initial point, will The absolute time that historical flood occurs is converted to sequential；

Step 3: select flood similarity indices, and each index is combined, form different flood similarity indices groups Conjunction scheme, described flood similar figures index has included rise flow, period accumulation rainfall and the period accumulation water yield；

Step 4: similar flood Dynamic Recognition, builds similar flood collection, and some for history floods are each considered as a class, calculates each Two classes that distance is minimum are merged into a new class, then under the differentiation of new classification, recalculate by the distance between class Distance between all kinds of, then two classes that distance is minimum are merged into a new class, until all patterns are polymerized to two classes；

Step 5: choose similar flood；Rise by flood section if being in or still have bigger follow-up rainfall, then concentrating from similar flood and select Last similar flood long, that magnanimity is big or take the outer envelope curve of similar flood collection；If being in the afterbody of rainfall, then from similar flood Water is concentrated and is selected last similar flood short, that magnanimity is little or take the interior envelope curve of similar flood collection；Other situations then select similar flood The averaging process of water collection；

Step 6: similar flood extension；In Real-time Flood Forecasting leading time, use the achievement of Real-time Flood Forecasting；At leading time Outward, the similar flood of selection is concentrated to splice from similar flood；

Wherein, the step building similar flood collection farther includes:

Step 41, initialization classification, choose the flood that history N field data is the most complete, calculate each field K similarity indices of flood Numerical value, is each considered as a class by history N field flood, is designated as: G_i ⁽⁰⁾={ x_iI=1,2 ..., N, N, K are natural number；

Step 42, calculate all kinds of between Euclidean distance D_i,j, generate a full symmetric Distance matrix D^(K)=(D_i,j)_m×m, Wherein, m is the number of class, m=N time initial；Euclidean distance D_i,jBelow equation can be used to calculate:

Q₀The flow that rises is acted for flood；P_tFor period accumulation rainfall；W_tFor the period accumulation water yield；

Matrix D in step 43, calculation procedure 42^(K)In least member, if it is G_a ^(K)And G_b ^(K)Between distance, by G_a ^(K)And G_b ^(K)Being merged into a class, the newest cluster is designated as: G₁ ^(K+1),G₂ ^(K+1)..., make k=k+1, m=m-1；

Step 44, the number of inspection class, such as number m of fruit > 2, go to step 42 and proceed cluster calculation；Otherwise, calculate Terminate；

Step 45, build similar flood collection, according to the result of Hierarchical clustering analysis, take front some floods and build similar flood Collection；

Described similar flood expanding step farther includes:

When flood splicing moment real-time prediction value is inconsistent to similar flood discharge value, use the similar flood of integral translation The method of water process realizes linking up smoothly, and below equation can be used to calculate:

Δ Q=Q (t₀+τ)-Q₁(t₀+τ)

Q₀The flow that rises is acted for flood；

t₀For current time；

τ is Real-time Flood Forecasting leading time；

Q (t) is the discharge process of Real-time Flood Forecasting；

Q₁T () is the similar flood discharge process chosen；

Q₂T () is similar flood extension discharge process；

Δ Q is t₀The difference of+τ moment real-time prediction flow value and similar flood discharge value.

2. the Application of Data Mining of the follow-up water of extension reservoir as claimed in claim 1, it is characterised in that also include step Rapid 7:

Evaluate flood extension effect；Flood extension effect is evaluated by follow-up water yield extension ratio of profit increase and deterministic coefficient increment；

Follow-up water yield extension ratio of profit increase uses below equation to calculate:

Wherein, Q_RealT () is observed flood process；T is schedule periods, t₀For current time, Q (t) is the flow mistake of Real-time Flood Forecasting Journey；Q₂T () is similar flood extension discharge process；

Deterministic coefficient increment uses below equation to calculate:

Δ DC=DC₂-DC

Wherein, DC₂Deterministic coefficient for similarity extension peb process；DC is the deterministic coefficient of real-time prediction peb process；

It is respectively adopted below equation to calculate:

Wherein,For the meansigma methods of observed flood process, T is schedule periods, t₀For current time.