CN103854055A - Electric transmission line icing prediction model based on neural network and fuzzy logic algorithm - Google Patents

Abstract

The invention discloses an electric transmission line icing prediction model based on a neural network and a fuzzy logic algorithm. The model comprises the following steps: reading micro meteorological parameters to form a training sample, modifying the weight of a network, introducing a threshold, acquiring the fundamental component of the icing thickness, reading position information of a pole tower, establishing an altitude subordinating degree function and a large area moisture distance subordinating degree function, establishing an error correction subordinating degree function, forming a fuzzy rule bank so as to obtain correction coefficients through defuzzification, and combining the calculation result of the neural network and the fuzzy logic compensation result. The electric transmission line icing prediction model with geographical location information is high in prediction precision when being compared with a conventional global model and a single BP (Back Propagation) neural network, and has a good effect in practical application.

Description

Powerline ice-covering forecast model based on neural network and fuzzy logic algorithm

Technical field

The present invention proposes a kind of powerline ice-covering combination forecasting based on neural network and fuzzy logic algorithm, belong to power system security protection field.

Background technology

Transmission line of electricity is as the artery of electric system, and for the cooperating operation between user power utilization and electrical network provides passage, its critical role is unquestionable.But, because circuit exposes field for a long time to the open air, not only exist unaccelerated aging, deteriorated problem, but also be a huge environmental hazard supporting body, become part the most fragile in electric system.Especially in the winter time, some regional line ice coatings are comparatively serious, cause steel tower distortion, the faults such as tower, broken string of falling happen occasionally, and cannot guarantee the sustainability of customer power supply, give produce, life causes serious impact.Meanwhile, also making to rush to repair operational difficulties degree increases.The icing disaster of therefore, how to prevent and treat transmission line of electricity become intelligent grid build in one of problem demanding prompt solution.It has also caused the extensive concern of Chinese scholars and scientific research institution.

By consulting lot of documents, although find that at present the total class of icing forecast model about power circuit is various, its essence is nothing more than being experimental formula based on mathematics or the icing model based on intelligent algorithm.

Wherein, Russia, Canada-United States, etc. the researchist of state line ice coating has been carried out to a large amount of research, a large amount of theoretical results and product have been obtained in the field such as mechanism, wire icing load at wire icing, but their research more be the mathematic(al) experience formula of line ice coating.But the method often lacks detailed Data support in actual application, and application and the expansion of restriction the method.

At home, each design, scientific research and run unit have also carried out a large amount of research work, University Of Chongqing has proposed the multiple icing model based on mathematical algorithm and intelligent algorithm, obtained many fruitful achievements, but their achievement in research seldom relates to the microclimate condition in powerline ice-covering and somewhere is combined.In other words, research and analyse be substantially on a large scale, the overall situation line ice coating model.But, fall the multiple location of tower at icing, be in most cases by due to the severe microclimate condition in this location, therefore world model is in this location and inapplicable.

Summary of the invention

The object of the invention is to: propose a kind of powerline ice-covering combination forecasting based on neural network and fuzzy logic algorithm, neural network is combined with fuzzy logic algorithm, fully in conjunction with microclimate parameter, the ice covering thickness of prediction transmission line of electricity, set up powerline ice-covering model, for the laying of transmission line of electricity provides reference, safeguard power system security.

Solution of the present invention is: a kind of powerline ice-covering forecast model based on neural network and fuzzy logic algorithm, it is characterized in that, and comprise the steps:

Step 1: read in the microclimate parameter of historical microclimate point, composing training sample; Microclimate parameter comprises environment temperature, ambient wind velocity, ambient humidity, lake distance, sea level elevation and actual ice thickness;

Step 2: determine BP neural network model parameter, utilize the weight w of data processing formula roll-off network _ij, introduce threshold value θ _j;

Step 3: complete neural network iterative process, obtain ice covering thickness fundametal component o _pi;

Step 4: read in shaft tower positional information; Shaft tower positional information comprises sea level elevation, large region steam distance, environment temperature, mean wind direction, ambient humidity.

Step 5: add the method for fuzzy logic compensation, based on geographic position factor, set up sea level elevation membership function and large region steam apart from membership function for sea level elevation and large region steam distance respectively;

Step 6: apart from membership function, set up error correction membership function according to sea level elevation membership function and large region steam, form fuzzy rule base and also draw correction coefficient by defuzzification; General large region steam refers to lake, the distance between large region steam criterion distance sampled point and lake;

Step 7, combines neural computing result and fuzzy logic compensation result, draws the powerline ice-covering combination forecasting based on neural network and fuzzy logic algorithm.

In step 2, BP neural network model comprises input layer, hidden layer and output layer; Described BP neural network model parameter comprises input layer, hidden layer neuron and output layer neuron; Described input layer comprises environment temperature, ambient wind velocity, ambient humidity and actual ice thickness, determines the meteorologic parameter input type of each microclimate point by historical input model; Described hidden layer neuron is the node of hidden layer, and the node of described hidden layer carries out nodes adjustment according to training precision in training process; Described output layer neuron is ice covering thickness h _oki;

The weight w of roll-off network _ijwith threshold value θ _jspecifically comprise the following steps:

The meteorologic parameter of supposing input has n, and hidden layer has m neuron, and the error calculation formula of k sample is formula (1):

$e_k=\frac{{(h_{\mathrm{ki}}-h_{\mathrm{oki}})}^2}{2}---\left(1\right)$

In formula: h _kifor the ice covering thickness output of expecting, neuron is output as ice covering thickness h _oki, e _kfor the error of k sample;

The weight w of BP neural network roll-off network _ijalgorithm is respectively formula (2) and formula (3):

$u_i=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{w}_{\mathrm{ij}}{x}_j-{\mathrm{\θ}}_j,(i=\mathrm{1,2},.....m)---\left(2\right)$

v _i＝f(u _i) （i=1，2，.....m) (3）

Threshold value θ _jset according to historical data;

In formula: f () is hidden layer input-output function, generally selects tansig function, x _jfor the meteorologic parameter that step 1 is inputted, u _ifor hidden layer node input, v _ifor hidden layer node output;

Introduce momentum λ for the speed of convergence of accelerating BP neural network, its formula is formula (4) and formula (5):

$\mathrm{\Δ}{w}_{\mathrm{ij}}^n{\left(t\right)}^{\′}=-\mathrm{\η}\frac{\∂e_k}{\∂w}=-\mathrm{\η}\frac{\∂e_k}{\∂h_{\mathrm{oki}}}\frac{\∂h_{\mathrm{oki}}}{\∂w}=\mathrm{\η}(h_{\mathrm{ki}}-h_{\mathrm{oki}})\frac{\∂h_{\mathrm{oki}}}{\∂w}---\left(4\right)$

$\mathrm{\Δ}{w}_{\mathrm{ij}}^n\left(t\right)=\mathrm{\Δ}{w}_{\mathrm{ij}}^n{\left(t\right)}^{\′}+\mathrm{\λ\Δ}{w}_{\mathrm{ij}}^n(t-1)---\left(5\right)$

In formula:

for j neuron of t moment n layer do not introduced the weight coefficient increment of momentum factor to i neuron of n+1 layer,

for the correction result in t moment,

for the correction result in t-1 moment, η is Ratio for error modification;

Back-pushed-type (6) is to former

carry out corrected Calculation and obtain new connection weight vector w _ij, new connection weight vector w _ijbe finally revised

$w_{\mathrm{ij}}^n\left(t\right)+\mathrm{\Δ}{w}_{\mathrm{ij}}^n\left(t\right)\→w_{\mathrm{ij}}^n\left(t\right)---\left(6\right)$

The above computation process, be along forward computation process oppositely, by control information transmission, to connection weight vector w _ijcarry out corrected Calculation.

In step 3 based on BP neural network iterative algorithm, by output layer neuron ice covering thickness h _okiafter revising, obtain ice covering thickness fundametal component o _pi, ice covering thickness fundametal component o _pifor output layer neuron is ice covering thickness h _okithrough the output after revising, ice covering thickness fundametal component o _picomputing formula is formula (7):

o _pi＝z(v _it _ij)（i=1,2,…..m） （7）

Z () represents output function, generally selects logsig function, t _ijfor the connection weight of hidden layer neuron i and output layer neuron j, v _ifor hidden layer node output, v _iobtain according to formula (2) and formula (3).

The large region of step 5 steam is set up apart from membership function, comprise the following steps: adopt 7 kinds of fuzzy language variate-value NB1, NM1, NS1, ZE1, PS1, PM1, PB1 to carry out obfuscation to large region steam distance, form 7 fuzzy sets, NB1, NM1, NS1, ZE1, PS1, PM1, PB1 represent respectively the large region of sampled point distance steam distance negative large, negative in, negative little, zero, just little, just neutralize honest;

More preferably, the large region of NB1 steam distance range is that the large region of 0m～30m, NM1 steam distance range is 10m～50m, the large region of NS1 steam distance range is 30m～90m, the large region of ZE1 steam distance range is 50m～110m, the large region of PS1 steam distance range is 90m～160m, the large region of PM1 steam distance range is 130m～200m, and the large region of PB1 steam distance range is for being greater than 160m.

Sea level elevation membership function is set up, comprise the following steps: adopt 4 kinds of fuzzy language variate-value AL, AM, AH, AU, the residing sea level elevation of circuit has been carried out to obfuscation, and wherein AL, AM, AH, AU have represented respectively basic, normal, high, high four differing heights of electric power line pole tower sea level elevation;

More preferably, the sea level elevation scope of AL is 0m～500m, and the sea level elevation scope of AM is 100m～1100m, and the sea level elevation scope of AH is 900m～1700m, and the sea level elevation scope of AU is 1400m～1800m.

Described in step 6, error correction membership function is set up and is comprised the following steps: get 5 fuzzy variables, NB2, NS2, ZE2, PS2, PB2 set up correction coefficient membership function; NB2, NS2, ZE2, PS2, PB2 represent line ice coating influence coefficient negative large, negative little, zero, just little and honest, described line ice coating influence coefficient is correction coefficient;

More preferably, NB2 represent that correction coefficient scope is-10～-5, NS2 represent correction coefficient scope for-10～0, ZE2 represent correction coefficient scope for-5～5, PS2 represents that correction coefficient scope is 0～10, PB2 represents that correction coefficient scope is 5～10; Above-mentioned correction coefficient is the optimal value that test of many times is obtained;

Step 6 fuzzy rule base is set up and is comprised the following steps, and sets up membership function and history run experience according to sea level elevation membership function, large region steam distance, adopts fuzzy condition statement if ... and ... the inference method of then, set up fuzzy rule base;

Simultaneously, the large region of foundation steam is apart from membership function and sea level elevation membership function, obtain the degree of membership of the large region of sampled point meteorologic parameter steam distance and sea level elevation, by described degree of membership substitution corrective system membership function, obtain two end points of degree of membership in fuzzy set, the center that obtains corresponding fuzzy set by two end points are got to intermediate value;

Finally adopting gravity model appoach to carry out defuzzification to fuzzy rule is formula (8):

Wherein

be the center of i fuzzy set, w _ibe the degree of membership of i fuzzy set, the degree of membership of described i fuzzy set comprises that the large region steam of i fuzzy set is apart from degree of membership and sea level elevation degree of membership, and M is fuzzy set number, and l is the error correction coefficient that fuzzy logic draws.

In step 7, the built-up pattern framework ice covering thickness formula of neural network and fuzzy logic is formula (9):

D _h＝o _pi+o _pi×l/100 (9)

In formula: D _hfor the ice covering thickness value of prediction, o _pifor revised ice covering thickness, be ice covering thickness fundametal component, l is the error correction coefficient that fuzzy logic draws.

Technique effect of the present invention:

1) the method is furtherd investigate for powerline ice-covering rule under the microclimate condition of mountain area, and obtain under different microclimate conditions, the influence coefficient difference such conclusions of each meteorologic parameter to powerline ice-covering, and because transmission line of electricity geographical conditions difference of living in is also not identical on the impact of powerline ice-covering.For this feature, the powerline ice-covering combination forecasting that this patent has proposed the powerline ice-covering prediction built-up pattern of combining geographic location information-based on neural network and fuzzy logic algorithm, has obtained good effect in actual applications;

2) in partial model of the present invention matching neural network, compare world model more targeted, calculate more excellent weights, predict the outcome more accurately;

3) fuzzy logic method can be good at compensating the error of neural network output;

4) powerline ice-covering generation is by multiple meteorological condition, the result of the non-linear strong coupling factor effects such as shaft tower geographic position, in order to include geographical environmental condition in limit of consideration, adopt fuzzy logic by information fuzzies such as geographic position, and utilize neural network to carry out matching to historical icing data and weather data, thereby set up line ice coating prediction built-up pattern, can find out from example of calculation shows, this model is compared BP neural network higher precision of prediction.

Accompanying drawing explanation

Fig. 1 powerline ice-covering forecast model schematic diagram based on neural network and fuzzy logic algorithm of the present invention;

Fig. 2 is neural network structure schematic diagram;

Fig. 3 is large domain type steam membership function;

Fig. 4 is sea level elevation membership function;

Fig. 5 is correction coefficient membership function;

The precision of prediction comparative result figure of Fig. 6 world model and partial model;

Error and this patent model comparison diagram that Fig. 7 does not adopt the icing forecast model of fuzzy logic method to produce.

Embodiment

Solution of the present invention is: a kind of powerline ice-covering combination forecasting based on neural network and fuzzy logic algorithm, the present embodiment has adopted ice covering thickness data, shaft tower geographical location information and the prediction of microclimate data analysis thereof of 2 microclimate points in certain mountain area, comprises the steps:

A powerline ice-covering forecast model based on neural network and fuzzy logic algorithm, is characterized in that as shown in Figure 1, comprises the steps:

S01: read in the microclimate parameter of historical microclimate point, composing training sample;

S02: determine BP neural network model parameter, utilize the weight w of the continuous roll-off network of data processing formula _ijwith threshold value θ _j;

S03: complete neural network iterative process, obtain ice covering thickness fundametal component o _pi;

S04: read in shaft tower positional information, comprise sea level elevation, with the distance of large region steam etc. on the darker geographical conditions information of icing impact;

S05: add the method for fuzzy logic compensation, consider geographic position factor, set up sea level elevation membership function and large region steam apart from membership function for sea level elevation and large region steam distance respectively;

S06: apart from membership function, set up error correction membership function according to sea level elevation membership function and large region steam, form fuzzy rule base and also draw correction coefficient by defuzzification;

S07, combines neural computing result and fuzzy logic compensation result, draws the powerline ice-covering combination forecasting based on neural network and fuzzy logic algorithm.

In S01, microclimate parameter comprises environment temperature, ambient wind velocity, ambient humidity, lake distance, sea level elevation and actual ice thickness, the present embodiment microclimate parameter (only enumerated the data of 4 sample number in the present embodiment, all the other sample datas are described no longer one by one) as shown in table 1.

Table 1 microclimate parameter

As shown in Figure 2, in S02, BP neural network model comprises input layer, hidden layer and output layer; Described BP neural network model parameter comprises input layer, hidden layer neuron and output layer neuron; Described input layer comprises environment temperature, ambient wind velocity, ambient humidity and actual ice thickness, determines the meteorologic parameter input type of each microclimate point by historical input model; Described hidden layer neuron is the node of hidden layer, and the node of described hidden layer carries out nodes adjustment according to training precision in training process; Described output layer neuron is ice covering thickness h _oki;

The meteorologic parameter of supposing input has n, and hidden layer has m neuron, and output layer neuron neuron is output as ice covering thickness h _oki, the error calculation formula of k sample is formula (1):

$e_k=\frac{{(h_{\mathrm{ki}}-h_{\mathrm{oki}})}^2}{2}---\left(1\right)$

In formula: h _kifor the ice covering thickness output of expecting, e _kfor the error of k sample;

The weight w of BP neural network roll-off network _ijalgorithm is respectively formula (2) and formula (3):

$u_i=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{w}_{\mathrm{ij}}{x}_j-{\mathrm{\θ}}_j,(i=\mathrm{1,2},.....m)---\left(2\right)$

v _i＝f(u _i) （i=1,2,…..m） （3）

Threshold value θ _jfor the optimal value of obtaining by historical data, the present embodiment threshold value θ _jsystem of selection with reference to fuzzy control and Intelligent Control Theory and application [M]. publishing house of Harbin Institute of Technology, 1991;

In formula: f () is hidden layer input-output function, select tansig function as hidden layer node input-output function, x _jfor the meteorologic parameter that step 1 is inputted, u _ifor hidden layer node input, v _ifor hidden layer node output;

Introduce momentum λ for the speed of convergence of accelerating BP neural network, its formula is formula (4) and formula (5) simultaneously:

$\mathrm{\Δ}{w}_{\mathrm{ij}}^n{\left(t\right)}^{\′}=-\mathrm{\η}\frac{\∂e_k}{\∂w}=-\mathrm{\η}\frac{\∂e_k}{\∂h_{\mathrm{oki}}}\frac{\∂h_{\mathrm{oki}}}{\∂w}=\mathrm{\η}(h_{\mathrm{ki}}-h_{\mathrm{oki}})\frac{\∂h_{\mathrm{oki}}}{\∂w}---\left(4\right)$

$\mathrm{\Δ}{w}_{\mathrm{ij}}^n\left(t\right)=\mathrm{\Δ}{w}_{\mathrm{ij}}^n{\left(t\right)}^{\′}+\mathrm{\λ\Δ}{w}_{\mathrm{ij}}^n(t-1)---\left(5\right)$

In formula: for j neuron of t moment n layer do not introduced the weight coefficient increment of momentum factor to i neuron of n+1 layer,

for correction result,

for the correction result in t-1 moment, η is Ratio for error modification;

Back-pushed-type (6) is to former

carry out corrected Calculation and obtain new connection weight vector w _ij, new connection weight vector w _ijbe repeatedly revised

$w_{\mathrm{ij}}^n\left(t\right)+\mathrm{\Δ}{w}_{\mathrm{ij}}^n\left(t\right)\→w_{\mathrm{ij}}^n\left(t\right)---\left(6\right)$

The above computation process, be along forward computation process oppositely, by control information transmission, to connection weight vector w _ijcarry out corrected Calculation.

In S03 based on BP neural network iterative algorithm, by output layer neuron ice covering thickness h _okiafter revising, obtain ice covering thickness fundametal component o _pi, ice covering thickness fundametal component o _picomputing formula is formula (7):

o _pi＝z(v _it _ij)（i=1,2,…m） （7）

Z () represents o _picorrection output function, select logsig function as ice covering thickness fundametal component o _picomputing formula, t _ijfor the connection weight of hidden layer neuron i and output layer neuron j, v _ifor hidden layer node output.

In S04, read in shaft tower positional information, comprise sea level elevation, large region steam distance (being lake distance), environment temperature, mean wind direction, ambient humidity, the present embodiment parameter (only enumerated the data of 4 sample number in the present embodiment, all the other sample datas are described no longer one by one) as shown in table 2.

Table 2 shaft tower positional information

As shown in Figure 3, the large region of S05 steam is set up (large region steam is lake apart from degree of membership apart from degree of membership) apart from membership function, comprise the following steps, adopt 7 kinds of fuzzy language variate-value NB1, NM1, NS1, ZE1, PS1, PM1, PB1 to carry out obfuscation to large region steam distance, form 7 fuzzy sets, NB1, NM1, NS1, ZE1, PS1, PM1, PB1 represent respectively the large region of sampled point distance steam distance negative large, negative in, negative little, zero, just little, just neutralize honest; The large region of described NB1 steam distance range is that the large region of 0m～30m, NM1 steam distance range is 10m～50m, the large region of NS1 steam distance range is 30m～90m, the large region of ZE1 steam distance range is 50m～110m, the large region of PS1 steam distance range is 90m～160m, the large region of PM1 steam distance range is 130m～200m, and the large region of PB1 steam distance range is for being greater than 160m; Large region steam is 0～1 apart from degree of membership span; By corresponding with degree of membership above-mentioned 7 kinds of fuzzy language variate-values, obtain large region steam apart from membership function.

As shown in Figure 4, sea level elevation membership function is set up, comprise the following steps: adopt 4 kinds of fuzzy language variate-value AL, AM, AH, AU, the residing sea level elevation of circuit has been carried out to obfuscation, wherein AL, AM, AH, AU has represented respectively the low of electric power line pole tower sea level elevation, in, high, high four differing heights, the sea level elevation scope of AL is 0m～500m, the sea level elevation scope of AM is 100m～1100m, the sea level elevation scope of AH is 900m～1700m, the sea level elevation scope of AU is 1400m～1800m, sea level elevation degree of membership value is 0～1, by AL, AM, AH, AU is corresponding with degree of membership, obtain sea level elevation membership function.

As shown in Figure 5, described in S06, the foundation of error correction membership function comprises the following steps: get 5 fuzzy variables, NB2, NS2, ZE2, PS2, PB2 set up correction coefficient membership function; NB2, NS2, ZE2, PS2, PB2 represent line ice coating influence coefficient negative large, negative little, zero, just little and honest, described line ice coating influence coefficient is correction coefficient; NB2 represents that correction coefficient scope is-10～-5, NS2 represent correction coefficient scope for-10～0, ZE2 represent correction coefficient scope for-5～5, PS2 represents that correction coefficient scope is 0～10, PB2 represents that correction coefficient scope is 5～10; Correction coefficient is the optimal value of obtaining by historical data.

Fuzzy rule base is set up and is comprised the following steps, set up membership function and history run experience according to sea level elevation membership function, large region steam distance, adopt fuzzy condition statement if ... and ... the inference method of then, set up fuzzy rule base, large domain type steam factor, sea level elevation can be taken into full account into the impact of powerline ice-covering.The wherein rule that for example the present embodiment adopts: if D is NM and Alt is AM then I is PS, wherein D representative is the distance from lake, and Alt is the residing sea level elevation of circuit, and I represents line ice coating influence coefficient.Singularity for each microclimate point line ice coating rule also can be done further sectionalization to error correction degree of membership.

As shown in Fig. 3, Fig. 4 and Fig. 5, in the present embodiment, in the time that sampled point is 20m apart from the distance in lake, its degree of membership that belongs to NB is 0.5, and the degree of membership that belongs to NM is 0.5, when sea level elevation is 200m, its degree of membership that belongs to AL is 1, and the degree of membership that belongs to AM is 0.2, when setting up rule: if distance=NB and height above sea level=AL then correction factor is PS, if distance=NM and height above sea level=AM then correction factor is ZE, time its to adopt gravity model appoach be shown in formula (8)

l=(0.5*5+7*0.5+-4*0.2+4*0.2)/(0.5+0.5+0.2+0.2)=4.28。

Built-up pattern framework ice covering thickness formula in described S07 is formula (9):

D _h＝o _pi+o _pi×l/100 (9)

In formula: D _hfor the ice covering thickness value of prediction, o _pifor ice covering thickness fundametal component (), l is the error correction coefficient that fuzzy logic draws.

Model predicts the outcome as (only enumerated the data of 4 sample number in the present embodiment, all the other sample datas are described no longer one by one) shown in table 3 after setting up.

The prediction of No. 50 ice covering thickness of table 3 shaft tower and actual contrast

For the ease of comparing, the present embodiment compares result as shown in Figure 6 by the precision of prediction of world model and partial model.As seen from Figure 6, due to neural network matching in partial model, to compare world model more targeted, thereby produced more excellent weights, and result is to predict the outcome more accurately.And do not adopt error and the present embodiment model that the icing forecast model of fuzzy logic method produces to contrast as shown in Figure 6.As shown in Figure 7, fuzzy logic method can be good at compensating the error of neural network output in some cases, but is causing occurring over-compensation under certain situation because modeling comparison is coarse, can in the time further studying from now on, improve the precision of modeling.

Below be only the preferred embodiment of the present invention; be noted that for those skilled in the art; under the premise without departing from the principles of the invention, can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.

Claims (10)

1. the powerline ice-covering forecast model based on neural network and fuzzy logic algorithm, is characterized in that, comprises the steps:

Step 1: read in the microclimate parameter of historical microclimate point, composing training sample;

Step 2: determine BP neural network model parameter, utilize the weight w of data processing formula roll-off network _ij, introduce threshold value θ _j;

Step 3: complete neural network iterative process, obtain ice covering thickness fundametal component o _pi;

Step 4: read in shaft tower positional information;

Step 5: add the method for fuzzy logic compensation, based on geographic position factor, set up sea level elevation membership function and large region steam apart from membership function for sea level elevation and large region steam distance respectively;

Step 6: apart from membership function, set up error correction membership function according to sea level elevation membership function and large region steam, form fuzzy rule base and also draw correction coefficient by defuzzification;

Step 7, combines neural computing result and fuzzy logic compensation result, draws the powerline ice-covering combination forecasting based on neural network and fuzzy logic algorithm.

2. the powerline ice-covering forecast model based on neural network and fuzzy logic algorithm according to claim 1, is characterized in that: the parameter of microclimate described in step 1 comprises environment temperature, ambient wind velocity, ambient humidity, lake distance, sea level elevation and actual ice thickness.

3. the powerline ice-covering forecast model based on neural network and fuzzy logic algorithm according to claim 1, is characterized in that:

In described step 2, BP neural network model comprises input layer, hidden layer and output layer; Described BP neural network model parameter comprises input layer, hidden layer neuron and output layer neuron; Described input layer comprises environment temperature, ambient wind velocity, ambient humidity and actual ice thickness, determines the meteorologic parameter input type of each microclimate point by historical input model; Described hidden layer neuron is the node of hidden layer, and the node of described hidden layer carries out nodes adjustment according to training precision in training process; Described output layer neuron is ice covering thickness h _oki.

4. the powerline ice-covering forecast model based on neural network and fuzzy logic algorithm according to claim 1, is characterized in that the weight w of described step 2 roll-off network _ijwith threshold value θ _jspecifically comprise the following steps:

The meteorologic parameter of supposing input has n, and hidden layer has m neuron, and the error calculation formula of k sample is formula (1):

$e_k=\frac{{(h_{\mathrm{ki}}-h_{\mathrm{oki}})}^2}{2}---\left(1\right)$

In formula: h _kifor the ice covering thickness output of expecting, neuron is output as ice covering thickness h _oki, e _kfor the error of k sample;

The weight w of BP neural network roll-off network _ijwith threshold value θ _jalgorithm is respectively formula (2) and formula (3):

$u_i=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{w}_{\mathrm{ij}}{x}_j-{\mathrm{\θ}}_j,(i=\mathrm{1,2},.....m)---\left(2\right)$

v _i＝f(u _i) （i=1,2,…..m） （3）

In formula: f () is hidden layer input-output function, x _jfor the meteorologic parameter that step 1 is inputted, u _ifor hidden layer node input, v _ifor hidden layer node output;

Introduce momentum λ for the speed of convergence of accelerating BP neural network, its formula is formula (4) and formula (5):

$\mathrm{\Δ}{w}_{\mathrm{ij}}^n{\left(t\right)}^{\′}=-\mathrm{\η}\frac{\∂e_k}{\∂w}=-\mathrm{\η}\frac{\∂e_k}{\∂h_{\mathrm{oki}}}\frac{\∂h_{\mathrm{oki}}}{\∂w}=\mathrm{\η}(h_{\mathrm{ki}}-h_{\mathrm{oki}})\frac{\∂h_{\mathrm{oki}}}{\∂w}---\left(4\right)$

$\mathrm{\Δ}{w}_{\mathrm{ij}}^n\left(t\right)=\mathrm{\Δ}{w}_{\mathrm{ij}}^n{\left(t\right)}^{\′}+\mathrm{\λ\Δ}{w}_{\mathrm{ij}}^n(t-1)---\left(5\right)$

In formula: for j neuron of t moment n layer do not introduced the weight coefficient increment of momentum factor to i neuron of n+1 layer,

for the correction result in t moment, for the correction result in t-1 moment, η is Ratio for error modification;

Back-pushed-type (6) is to former

carry out corrected Calculation and obtain new connection weight vector w _ij, described new connection weight vector w _ijbe through repeatedly revised

$w_{\mathrm{ij}}^n\left(t\right)+\mathrm{\Δ}{w}_{\mathrm{ij}}^n\left(t\right)\→w_{\mathrm{ij}}^n\left(t\right)---\left(6\right)$

The above computation process, be along forward computation process oppositely, by control information transmission, to connection weight vector w _ijcarry out corrected Calculation.

5. the powerline ice-covering forecast model based on neural network and fuzzy logic algorithm according to claim 3, is characterized in that: in described step 3 based on BP neural network iterative algorithm, by output layer neuron ice covering thickness h _okiafter revising, obtain ice covering thickness fundametal component o _pi, ice covering thickness fundametal component o _picomputing formula is formula (7):

o _pi＝z(v _it _ij) （i=1,2,…..m） （7）

Z () represents ice covering thickness fundametal component o _pioutput function, t _ijfor the connection weight of hidden layer neuron i and output layer neuron j, v _ifor hidden layer node output.

6. the powerline ice-covering forecast model based on neural network and fuzzy logic algorithm according to claim 1 is characterized in that: in described step 4, shaft tower positional information comprises sea level elevation, large region steam distance, environment temperature, mean wind direction, ambient humidity.

7. the powerline ice-covering forecast model based on neural network and fuzzy logic algorithm according to claim 1, is characterized in that,

The large region of step 5 steam is set up apart from membership function, comprise the following steps: adopt 7 kinds of fuzzy language variate-value NB1, NM1, NS1, ZE1, PS1, PM1, PB1 to carry out obfuscation to large region steam distance, form 7 fuzzy sets, described NB1, NM1, NS1, ZE1, PS1, PM1, PB1 represent respectively the large region of sampled point distance steam distance negative large, negative in, negative little, zero, just little, just neutralize honest;

Sea level elevation membership function is set up, comprise the following steps: adopt 4 kinds of fuzzy language variate-value AL, AM, AH, AU, the residing sea level elevation of circuit has been carried out to obfuscation, and wherein AL, AM, AH, AU represent respectively basic, normal, high, high four differing heights of electric power line pole tower sea level elevation.

8. the powerline ice-covering forecast model based on neural network and fuzzy logic algorithm according to claim 1, it is characterized in that, described in step 6, error correction membership function is set up and is comprised the following steps: get 5 fuzzy variables, NB2, NS2, ZE2, PS2, PB2 set up correction coefficient membership function; NB2, NS2, ZE2, PS2, PB2 represent line ice coating influence coefficient negative large, negative little, zero, just little and honest.

9. the powerline ice-covering forecast model based on neural network and fuzzy logic algorithm according to claim 1, is characterized in that,

Step 6 fuzzy rule base is set up and is comprised the following steps, and sets up membership function and history run experience according to sea level elevation membership function, large region steam distance, adopts fuzzy condition statement if ... and ... the inference method of then, set up fuzzy rule base;

The large region of foundation steam is apart from membership function and sea level elevation membership function, obtain the degree of membership of the large region of sampled point meteorologic parameter steam distance and sea level elevation, by described degree of membership substitution corrective system membership function, obtain two end points of degree of membership in fuzzy set, the center that obtains corresponding fuzzy set by two end points are got to intermediate value;

Finally adopting gravity model appoach to carry out defuzzification to fuzzy rule is formula (8):

Wherein

be the center of i fuzzy set, w _ibe the degree of membership of i fuzzy set, the degree of membership of described i fuzzy set comprises that the large region steam of i fuzzy set is apart from degree of membership and sea level elevation degree of membership, and M is fuzzy set number, and l is the error correction coefficient that fuzzy logic draws.

10. the powerline ice-covering forecast model based on neural network and fuzzy logic algorithm according to claim 1, is characterized in that,

In described step 7, the built-up pattern framework ice covering thickness formula of neural network and fuzzy logic is formula (9):

D _h＝o _pi+o _pi×l/100 (9)

In formula: D _hfor the ice covering thickness value of prediction, o _pifor revised ice covering thickness fundametal component, l is the error correction coefficient that fuzzy logic draws.

Images