CN105184421A - Electromagnetic environment parameter prediction method based on data segmentation and model calibration - Google Patents

Abstract

The invention relates to an electromagnetic environment parameter prediction method based on data segmentation and model calibration. The method belongs to the technical field of power system electromagnetic environment protection. The method mainly comprises: randomly dividing data obtained from measurement of an electromagnetic environment of a power transmission line into three blocks of the same size; conducting model fitting of different independent variable combinations on the first block of data and obtaining the independent variable combination with the minimum prediction error and a preliminary prediction model; applying the preliminary prediction model into the second block of data for correction, correcting each coefficient and constant of the model, and obtaining a prediction model to be verified; and finally, applying the to-be-verified model to the third block of data for testing. If the final prediction error and other statistical test indices satisfy requirements, the model is applicable for electromagnetic environment prediction of the power transmission line. The method can predict power transmission line electromagnetic environment parameters conveniently, effectively and reliably. The prediction results can be applied to many fields such as power transmission and transformation engineering design, optimization, electromagnetic environment protection, etc.

Description

Based on the electromagnetic environment parameter Forecasting Methodology of data sectional and model calibration

Technical field

The invention belongs to Power System Electromagnetic Environment guard technology field, particularly in order to predict the data analysing method of high voltage power transmission and transforming system electromagnetic environment parameter.

Background technology

Electromagnetic environment is the major consideration of high voltage power transmission and transforming system, and the root of the electromagnetic environment of transmission line of electricity is the corona discharge of circuit.Corona discharge can cause electric energy loss on the one hand, increases Transmission Cost; Affect wire periphery electromagnetic environment on the other hand, further the normal life of interference people.Along with the development of economy and the enhancing of common people's environmental consciousness, electromagnetic environment problem is further noticeable, and its electromagnetic environment problem of ultra-high-tension power transmission line has become the main restricting factor of its system and operation.The electromagnetic environment parameter of ultra-high-tension power transmission line mainly includes audible noise, radio interference, ground formate field intensity, ground ion flow density and corona loss.If in circuit design phase reasonable prediction electromagnetic environment parameter, then can optimize further line parameter circuit values such as wire line style, wire heading spacing, die openings, realize more eco-friendly power network development mode; Or under the rational electromagnetic environment level of guarantee, reduce transmission of electricity corridor as far as possible, reduce circuit expropriation of land and construction cost.This is to build and high-voltage testing room in planning has Important Project practical value.

Electromagnetic environment parameter influence factor is numerous, and has obvious random character.In electromagnetic environment parameter Forecasting Methodology in the past, utilize the electromagnetic environment data tested and obtain, take line parameter circuit value as independent variable, to the direct single matching of data, obtain electromagnetic environment parameter model of fit and predict for electromagnetic environment parameter.In this kind of method, data use and precision and reliability not fully can be caused all not high.And in this method without independent variable analysis and corrected model parameter process, under the result obtained can not reflect this environment completely, meteorologic parameter is on the impact of electromagnetic environment.Therefore electromagnetic environment parameter Forecasting Methodology in the past does not have sufficiently high engineering practical value.

Summary of the invention

The object of the invention is the weak point for overcoming prior art, proposing a kind of electromagnetic environment parameter Forecasting Methodology based on data sectional and model calibration; This method has been carried out classifying rationally to electromagnetic environment data and has been made full use of, and has carried out model calibration to EME forecast model, improves its precision of prediction; Forecast model is verified simultaneously, improve the engineering practical value of this model.

The electromagnetic environment parameter Forecasting Methodology based on data sectional and model calibration that the present invention proposes, it is characterized in that, the method specifically comprises the following steps:

1) the deblocking stage: electromagnetic environment parameter (audible noise will be comprised to the actual data measured of transmission line of electricity electromagnetic environment, radio interference, ground formate field intensity, ground ion flow density and corona loss), meteorologic parameter (wind speed, wind direction, air pressure, rainfall, temperature, relative humidity), line parameter circuit value (die opening, minimum height of conductor above ground, heading spacing) Data_base is divided into three pieces of subdata collection described by formula (1) at random: fitting data Data_trainning, correction data Data_calibration, verification msg Data_testing subdata collection, the data that these three pieces of subdatas are concentrated are respectively used to models fitting, parameter correction and modelling verification stage,

$\left\{\begin{array}{c}Data\_trainning=\{y_i,x_{i1},x_{i2},...,x_{im}\},i=1,2,...,n/3\\ Data\_calibration=\{y_j,x_{j1},x_{j2},...,x_{jm}\},j=n/3+1,n/3+2,...,2n/3\\ Data\_testing=\{y_k,x_{k1},x_{k2},...,x_{km}\},k=2n/3+1,2n/3+2,...,n\end{array}\right.---\left(1\right)$

In above formula, n is total data volume; Y is the one in electromagnetic environment parameter; X is the set of meteorologic parameter and line parameter circuit value, and wherein m is parameter kind sum;

2) the models fitting stage: will the array configuration of independent variable and applicable Statistical Prediction Model be determined; To the data in first piece of matching subdata set Data_trainning, utilize Linear Statistical Model to carry out matching according to the combination of different independents variable, wherein independent variable combination is first for comprising all variable x in the set of matching subdata _i1, x _i2..., x _im, be secondly the combination reducing independent variable number gradually, ensure that in this combination, meteorologic parameter and line parameter circuit value at least respectively have one simultaneously, obtain the tentative prediction model of shape such as formula (2), wherein, x' is a certain independent variable combination x _i1, x _i2..., x _ip, c ' is corresponding matrix of coefficients, the constant term that C ' is model of fit:

y′＝c′×x′+C′(2)

Obtain under different independent variable combines with corresponding matrix of coefficients based on formula (2), the error of electromagnetic environment parameter predicted value and actual value wherein, Y is actual value, for predicted value, get corresponding independent variable combination is as such as formula the tentative prediction model shown in (3):

y′＝c ₁′x ₁+c ₂′x ₂+…+c _p′x _p+C′(3)；

3) the parameter correction stage: according to step 2) the independent variable combination substitution second block correction data subset that obtains closes Data_calibration and carries out models fitting and obtain calibration model y "; Wherein, x " is independent variable combination x corresponding during the second block correction data subset closes _j1, x _j2..., x _jp, the matrix of coefficients that c " for x " is corresponding, C " is random entry;

y″＝c″×x″+C″(4)

Rudimentary model (3) and calibration model (4) subtract each other acquisition error (5), wherein Δ y=y '-y ", Δ c=c '-c ", Δ C=C '-C ";

Δy＝Δc ₁x ₁+Δc ₂x ₂+…+Δc _mx _m+ΔC(5)

Planning solution is adopted to solve linear programming problem (6)

minΔy＝Δc ¹x ¹+Δc ²x ²+…+Δc _px _p+ΔC

(6)

s.t.x∈{x ₁,x ₂,…,x _p}

To each coefficient in model and constant term c ₁', c ₂' ..., c _n', C ' carries out revising and obtains new coefficient and constant term c ₁, c ₂..., c _n, C, shown in (7), obtains model to be verified such as formula shown in (8);

<c ₁,c ₂,…,c _n,C>＝<c ₁′,c ₂′,…,c _n′,C′>+<Δc ₁,Δc ₂…Δc _p,ΔC>(7)

y＝c ₁x ₁+c ₂x ₂+…+c _nx _n+C(8)

4) the modelling verification stage: by step 3) in obtain correction after model to be verified substitute into the 3rd piece of verification msg subdata set Data_testing carry out test acquisition forecast model, if use the final predicated error of this forecast model be not more than step 2) in maximal value, then this forecast model is applicable to the EME forecast that experimental data obtains regional transmission line of electricity; Otherwise, return step 2) and re-start the models fitting stage.

Technical characterstic of the present invention and beneficial effect:

For the feature that electromagnetic environment data randomness is strong, carry out segmentation to data, make the precision of forecast model higher, the use of data is more abundant; For the feature that electromagnetic environment data influence factor is many, use different independent variable combinations to carry out matching respectively, obtain more rational EME forecast model independent variable; In model fitting process, add coefficient makeover process, the impact that reduction data randomness and improper data are brought, improve model accuracy.The electromagnetic environment model finally obtained can convenient, effectively, reliably carry out the prediction of transmission line of electricity electromagnetic environment parameter, predict the outcome and can be applied to multiple field such as project of transmitting and converting electricity Design and optimization and electromagnetic environment protection according to different demand, to the construction efficiency of project of transmitting and converting electricity be significantly improved, reducing the construction costs.

Accompanying drawing explanation

Fig. 1 is method flow block diagram of the present invention;

Embodiment

The electromagnetic environment parameter statistical analysis technique based on data sectional and model calibration that the present invention proposes is described with reference to the accompanying drawings embodiment.

Electromagnetic environment parameter Forecasting Methodology based on data sectional and model calibration of the present invention, as shown in Figure 1, the method specifically comprises the following steps:

1) the deblocking stage: will to the actual data measured of transmission line of electricity electromagnetic environment, comprise electromagnetic environment parameter (audible noise, radio interference, ground formate field intensity, ground ion flow density and corona loss), meteorologic parameter (wind speed, wind direction, air pressure, rainfall, temperature, relative humidity), line parameter circuit value (die opening, minimum height of conductor above ground, heading spacing) Data_base is divided into three pieces of subdata collection described by formula (1) at random: fitting data Data_trainning, correction data Data_calibration, verification msg Data_testing subdata collection, the data that these three pieces of subdatas are concentrated are respectively used to models fitting, parameter correction and modelling verification stage,

$\left\{\begin{array}{c}Data\_trainning=\{y_i,x_{i1},x_{i2},...,x_{im}\},i=1,2,...,n/3\\ Data\_calibration=\{y_j,x_{j1},x_{j2},...,x_{jm}\},j=n/3+1,n/3+2,...,2n/3\\ Data\_testing=\{y_k,x_{k1},x_{k2},...,x_{km}\},k=2n/3+1,2n/3+2,...,n\end{array}\right.---\left(1\right)$

In above formula, n is total data volume; Y is the one (predicting that a certain parameter then uses the data of this parameter) in electromagnetic environment parameter; X is the set of meteorologic parameter and line parameter circuit value, and wherein m is parameter kind sum;

2) the models fitting stage: will the array configuration of independent variable and applicable Statistical Prediction Model be determined; To the data in first piece of matching subdata set Data_trainning, (the present embodiment adopts conventional linear assembly language to utilize Linear Statistical Model according to the combination of different independents variable, LinearMix-effectedModel) carry out matching, wherein independent variable combination is first for comprising all variable x in the set of matching subdata _i1, x _i2..., x _im, be secondly the combination reducing independent variable number gradually, ensure that in combination, meteorologic parameter and line parameter circuit value at least respectively have one simultaneously, obtain the tentative prediction model of shape such as formula (2), wherein, x' is a certain independent variable combination x _i1, x _i2..., x _ip, c ' is corresponding matrix of coefficients, the constant term that C ' is model of fit:

y′＝c′×x′+C′(2)

Obtain under different independent variable combines with corresponding matrix of coefficients based on formula (2), the error of electromagnetic environment parameter predicted value and actual value wherein, Y is actual value, for predicted value, get corresponding independent variable combination is as such as formula the tentative prediction model shown in (3):

y′＝c ₁′x ₁+c ₂′x ₂+…+c _p′x _p+C′(3)；

3) the parameter correction stage: according to step 2) the independent variable combination substitution second block correction data subset that obtains closes Data_calibration and carries out models fitting and obtain calibration model y "; Wherein, x " is independent variable combination x corresponding during the second block correction data subset closes _j1, x _j2..., x _jp, the matrix of coefficients that c " for x " is corresponding, C " is random entry;

y″＝c″×x″+C″(4)

Rudimentary model (3) and calibration model (4) subtract each other acquisition error (5), wherein Δ y=y '-y ", Δ c=c '-c ", Δ C=C '-C ";

Δy＝Δc ₁x ₁+Δc ₂x ₂+…+Δc _mx _m+ΔC(5)

The conventional planning solutions such as simplicial method are adopted to solve linear programming problem (6)

minΔy＝Δc ¹x ¹+Δc ²x ²+…+Δc _px _p+ΔC

(6)

s.t.x∈{x ₁,x ₂,…,x _p}

To each coefficient in model and constant term c ₁', c ₂' ..., c _n', C ' carries out revising and obtains new coefficient and constant term c ₁, c ₂..., c _n, C, shown in (7), obtains model to be verified such as formula shown in (8);

<c ₁,c ₂,…,c _n,C>＝<c ₁′,c ₂′,…,c _n′,C′>+<Δc ₁,Δc ₂…Δc _p,ΔC>(7)

y＝c ₁x ₁+c ₂x ₂+…+c _nx _n+C(8)

4) the modelling verification stage: by step 3) in obtain correction after model to be verified substitute into the 3rd piece of verification msg subdata set Data_testing carry out test acquisition forecast model, if use the final predicated error of this forecast model be not more than step 2) in maximal value, then this forecast model is applicable to the EME forecast that experimental data obtains regional transmission line of electricity.Otherwise, return step 2) and re-start the models fitting stage.

Claims (1)

1., based on an electromagnetic environment parameter statistical analysis technique for data sectional and model calibration, it is characterized in that, the method specifically comprises the following steps:

1) the deblocking stage: basic data Data_base will be recorded to transmission line of electricity electromagnetic environment is actual: electromagnetic environment parameter, meteorologic parameter, line parameter circuit value are divided into three pieces of subdata collection described by formula (1) at random: fitting data Data_trainning, correction data Data_calibration, verification msg Data_testing subdata collection; The data that these three pieces of subdatas are concentrated are respectively used to models fitting, parameter correction and modelling verification stage;

$\left\{\begin{array}{c}Data\_trainning=\{y_i,x_{i1},x_{i2},...,x_{im}\},i=1,2,...,n/3\\ Data\_calibration=\{y_j,x_{j1},x_{j2},...,x_{jm}\},j=n/3+1,n/3+2,...,2n/3\\ Data\_testing=\{y_k,x_{k1},x_{k2},...,x_{km}\},k=2n/3+1,2n/3+2,...,n\end{array}\right.---\left(1\right)$

In above formula, n is total data volume; Y is the one in electromagnetic environment parameter; X is the set of meteorologic parameter and line parameter circuit value, and wherein m is parameter kind sum;

2) the models fitting stage: will the array configuration of independent variable and applicable Statistical Prediction Model be determined; To the data in first piece of matching subdata set Data_trainning, utilize Linear Statistical Model to carry out matching according to the combination of different independents variable, wherein independent variable combination is first for comprising all variable x in the set of matching subdata _i1, x _i2..., x _im, be secondly the combination reducing independent variable number gradually, ensure that in this combination, meteorologic parameter and line parameter circuit value at least respectively have one simultaneously, obtain the tentative prediction model of shape such as formula (2), wherein, x ' is a certain independent variable combination x _i1, x _i2..., x _ip, c ' is corresponding matrix of coefficients, the constant term that C ' is model of fit:

y′＝c′×x′+C′(2)

Obtain under different independent variable combines with corresponding matrix of coefficients based on formula (2), the error of electromagnetic environment parameter predicted value and actual value wherein, Y is actual value, for predicted value, get corresponding independent variable combination is as such as formula the tentative prediction model shown in (3):

y′＝c ₁′x ₁+c ₂′x ₂+…+c _p′x _p+C′(3)；

3) the parameter correction stage: according to step 2) the independent variable combination substitution second block correction data subset that obtains closes Data_calibration and carries out models fitting and obtain calibration model y "; Wherein, x " is independent variable combination x corresponding during the second block correction data subset closes _j1, x _j2..., x _jp, the matrix of coefficients that c " for x " is corresponding, C " is random entry;

y″＝c″×x″+C″(4)

Rudimentary model (3) and calibration model (4) subtract each other acquisition error (5), wherein, Δ y=y '-y ", Δ c=c '-c ", Δ C=C '-C ";

Δy＝Δc ₁x ₁+Δc ₂x ₂+…+Δc _mx _m+ΔC(5)

Simplicial method conventional planning solution is adopted to solve linear programming problem (6)

minΔy＝Δc ₁x ₁+Δc ₂x ₂+…+Δc _px _p+ΔC(6)

s.t.x∈{x ₁,x ₂,…,x _p}

To each coefficient in model and constant term c ₁', c ₂' ..., c _n', C ' carries out revising and obtains new coefficient and constant term c ₁, c ₂..., c _n, C, shown in (7), obtains model to be verified such as formula shown in (8);

<c ₁,c ₂,…,c _n,C>＝<c ₁′,c ₂′,…,c _n′,C′>+<Δc ₁,Δc ₂…Δc _p,ΔC>(7)

y＝c ₁x ₁+c ₂x ₂+…+c _nx _n+C(8)

4) the modelling verification stage: by step 3) in obtain correction after model to be verified substitute into the 3rd piece of verification msg subdata set Data_testing carry out test acquisition forecast model, if use the final predicated error of this forecast model be not more than step 2) in maximal value, then this forecast model is applicable to the EME forecast that experimental data obtains regional transmission line of electricity; Otherwise, return step 2) and re-start the models fitting stage.