A distribution network gale disaster early warning method based on PCA model

Abstract

The present invention belongs to the technical field of power grid disaster prevention, and specifically relates to a distribution network gale disaster early warning method based on a PCA model, comprising the following steps: according to a historical fault sample library, obtaining the entity parameters, refined meteorological parameters and operating state parameters when the fault occurs, forming a training sample set; using a multivariate regression method to fit the power outage duration of the training sample set, and establishing a power outage duration prediction model based on multivariate regression; using a PCA model to obtain the principal components of relevant multivariate parameters in the multivariate regression training process, and extracting the principal component parameters, and establishing a distribution network gale disaster power outage duration prediction model based on PCA; performing SVM classification on the distribution network gale disaster power outage duration; based on the prediction model and SVM classification obtained by training, early warning of gale disasters in the distribution network. The early warning method of the present invention realizes gale disaster early warning based on statistical methods, and is suitable for disaster prevention and mitigation of distribution networks with various complex situations.

Description

Technical Field

[0001] The invention belongs to the technical field of power grid disaster prevention, and in particular relates to a distribution network gale disaster early warning method based on a PCA model. Background Art

[0002] The power system is a huge and complex system involving multiple links such as generation, transmission, transformation, distribution and use. Its safe operation is closely related to the meteorological environment. Among them, the distribution network is the main object vulnerable to strong winds. Statistical data show that under the influence of strong winds, distribution network poles, broken wires, and hardware damage accidents frequently occur, and their severity is far greater than the impact of strong winds on the main grid's power transmission and transformation equipment. At the same time, strong winds often lead to foreign objects in the wind and the fall of surrounding trees caused by strong winds, which not only affects the safe and stable operation of the power grid, but also puts forward high requirements for emergency repairs and rapid power restoration. Therefore, under the premise of the stable demand for power load, the urgency of solving the above problems has been significantly enhanced, and it is urgent to propose an early warning method for strong wind disasters in the distribution network.

[0003] However, in the field of meteorology, accurate prediction of strong winds is a global problem. Since the distribution network focuses on a smaller scale, it is even more difficult to predict and warn of fine-grained strong wind disasters in the distribution field. At present, the strong wind disaster warning technology for the distribution network is still in the exploratory stage and is far from being able to support production applications. Power companies also lack effective strong wind monitoring methods and key data sources for early warning and forecasting with sufficiently fine granularity, and cannot form an effective distribution network strong wind disaster warning method. As a result, the distribution network wind disaster warning is not targeted, and the granularity of the disaster area locking range is relatively coarse, making it difficult to support the safe production, emergency response and material allocation of the distribution network.

[0004] At present, in terms of gale forecasting and early warning technology, professional meteorological agencies have initially formed hierarchical gale forecasts through years of development and upgrading. At the same time, combined with the continuously developing mesoscale numerical weather forecast system, they have studied the 3km*3km refined meteorological numerical forecast, combined with downscaling technology, formed a 1km*1km downscaling forecast for gale forecasts, and also formed minute-level gust forecasts. Although there has been a significant improvement in gale forecasting, it still cannot meet the production support needs of gale disaster early warning for small-scale distribution networks. It is very necessary to combine the existing meteorological forecast results, based on the historical distribution network gale fault disaster samples, use statistical methods, and combine the distribution network ledger data and operation status data to realize the early warning of gale disasters in the distribution network. Summary of the invention

[0005] In order to solve the deficiencies in the prior art, the present invention aims to provide a distribution network gale disaster early warning method based on the PCA model.

[0006] To achieve the above object, the present invention adopts the following technical solution:

[0007] A distribution network gale disaster early warning method based on a PCA model comprises the following steps:

[0008] Step 1: Obtain the main parameters, meteorological parameters, and operating status parameters of the distribution network under the gale fault disaster, and establish a distribution network historical gale fault sample library;

[0009] Step 2: Match the entity parameters, refined meteorological parameters and operating status parameters when the fault occurs to form a training sample set;

[0010] Step 3: Use the multiple regression method to fit the power outage duration of the training sample set in step 2 and establish a power outage duration prediction model based on multiple regression;

[0011] Step 4: Based on step 3, the PCA model is used to obtain the principal components of the relevant multivariate parameters in the multivariate regression training process, and the principal component parameters are extracted to establish a distribution network gale disaster power outage duration prediction model based on PCA;

[0012] Step 5: Based on step 3 and combined with the SVM algorithm, the duration of power outages caused by heavy wind disasters in the distribution network is classified by SVM according to the duration of power outages. According to the classification results, the level of heavy wind disasters based on different power outage durations is divided;

[0013] Step 6: In the prediction stage, according to the input distribution network parameters, refined meteorological parameters and operating status parameters, based on the prediction model trained in step 4, the power outage duration of different distribution network lines under strong wind disasters is predicted. Combined with the strong wind disaster level classification in step 5, the early warning of strong wind disasters in the distribution network is finally realized.

[0014] Preferably, the ontological parameter acquisition method described in step 1 is: statistically classifying the internal factors that affect the excitation of strong wind disasters in the distribution network, performing ontological parameter analysis, and obtaining the ontological parameters that affect the strong wind disasters in the distribution network.

[0015] Preferably, the main body parameters include tower information and line information of the distribution network.

[0016] Preferably, the tower information includes: tower type, number, longitude, latitude, altitude, manufacturer, installation location, geological environment information, horizontal span, vertical span, tower insulator string model, number of strings, number of insulator pieces, and tension tower rotation angle.

[0017] Preferably, the line information includes: line serial number, voltage level, line number, line name, starting and ending locations, line specifications, line type, number of loops, transmission length, design wind speed, design ice thickness, split number and split gap.

[0018] Preferably, the meteorological parameter acquisition method described in step 1 is: statistically classifying the external factors stimulated by the strong wind disaster that affects the operation of the distribution network, performing meteorological parameter analysis, and obtaining the meteorological parameters that affect the strong wind disaster in the distribution network.

[0019] Preferably, the meteorological parameters include: wind speed, wind direction, line angle, Doppler radar original parameters, whether there is precipitation, and precipitation amount.

[0020] Preferably, the steps for establishing the power outage duration prediction model based on multivariate regression described in step 3 are as follows:

[0021] 3.1 Input distribution network gale fault disaster training sample set X:

[0022] X={(X1,y1),(X2,y2),...(X i ,y i ),(X N ,y N )}

[0023] Among them, X i =(x i1 ,x i2 ,...x iM ), X i is the input parameter of the ith gale disaster fault sample, i.e., the distribution network main parameters, meteorological parameters and operation status parameters when the ith gale disaster fault sample occurs, x iM is the Mth parameter in the i-th gale disaster fault sample, y i is the power outage duration when the i-th gale disaster fault sample occurs;

[0024] 3.2 X i Substituting into the multiple regression formula, we get:

[0025] a1x i1 +a2x i2 +...a j x ij +a M x iM +b=y i

[0026] Among them, x ij is the jth parameter in the i-th gale disaster fault sample. Each sample has M parameters, a j is the weight to be trained corresponding to the jth parameter, and b is the constant to be trained;

[0027] Through multivariate regression training, we get the corresponding a j and b.

[0028] Preferably, the steps of establishing the PCA-based distribution network gale disaster power outage duration prediction model in step 4 are:

[0029] 4.1 According to the weights (a1, a2, ... a j ...,a M ,b), substitute into the PCA model and get:

[0030]

[0031] in, for The transposed matrix of , B is the PCA principal component model conversion matrix, and we get:

[0032] Where N<M, is the weight of the jth principal component parameter, is the constant obtained through training;

[0033] 4.2 Establish a PCA-based distribution network wind disaster power outage duration prediction model:

[0034]

[0035] y i ' is the predicted duration of power outage.

[0036] Preferably, the step of classifying the gale disaster levels based on different power outage durations described in step 5 is:

[0037] 5.1 Let φ(X) represent the feature vector after mapping X to high-dimensional space. The hyperplane model of SVM classifier can be expressed as:

[0038] f(X)=w T φ(X)+b

[0039] In the formula, f(X) is the hyperplane model, the training sample set X is the input, w is the weight, and b is the bias vector;

[0040] 5.2 Minimizing the Function

[0041]

[0042] stn i f(X i )≥1(i=1,2,…,m)

[0043] Where n i For X i Type label, X i is the input parameter of the i-th gale disaster fault sample;

[0044] 5.3 Since the formula in step 5.2 is a convex quadratic programming problem, adding the Lagrangian operator to it gives:

[0045]

[0046]

[0047] Introducing the kernel function k(X i ,X j ),satisfy:

[0048] k(X i ,X j )≤φ(X i ),φ(X j )≥φ(X i ) T φ(X j )

[0049] And the function is obtained as:

[0050]

[0051] Where: X i is the input parameter of the i-th gale disaster fault sample, X j is the input parameter of the jth gale disaster fault sample, n i For X i The type label of k(X i ,X j ) is the kernel function selected by SVM, α i is the Lagrangian operator;

[0052] 5.4 According to the SVM multi-classification strategy, select D main parameters and construct A binary classifier is used, and each classifier is trained for only two parameters. The distribution network gale disaster is divided into D levels according to the power outage duration.

[0053] Preferably, the distribution network gale disaster is divided into four levels according to the duration of power outage, namely: extremely minor power outage accident (≤0.5h), minor accident (0.5<t≤2h), general power outage accident (2h<t≤10h) and major power outage accident (>10h).

[0054] Preferably, the early warning steps for the gale disaster in the distribution network described in step 6 are:

[0055] 6.1X c =(x c1 ,x c2 ,...x cN ) is the principal component parameter of the cth case in the prediction stage, x cNis the Nth principal component parameter of the cth case, and is substituted into the prediction model trained in step 4 to obtain:

[0056] Get y c '=t

[0057] Among them, t is the predicted duration;

[0058] 6.2 Combined with the classification of gale disaster levels in step 5, the early warning of gale disasters in the distribution network can be finally achieved.

[0059] The positive beneficial effects of the present invention are:

[0060] 1. The present invention matches historical gale faults with meteorological information, distribution network parameter information and distribution network operating status information at the time of the fault occurrence through statistical analysis of the historical gale disaster fault set of the distribution network, forms a training sample set, performs multivariate regression, and fits the distribution curve of the gale fault of the distribution network. The PCA model is further used to extract the parameters of the principal components. According to the regression prediction, the most relevant principal component parameter weights are obtained, and a power outage duration prediction model of the distribution network under gale disasters is further obtained. According to the power outage duration caused by historical gale disasters, the SVM algorithm is used to divide the wind disasters into several levels. Finally, according to the prediction model obtained by training, the power outage duration when gale processes occur in the future can be predicted, the gale disaster level of the distribution network can be divided, and the strong convective gale disaster of the distribution network can be warned. The gale disaster warning based on statistical methods is realized, which is suitable for disaster prevention and mitigation of distribution networks with complex situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 This is a flow chart of the distribution network gale disaster early warning method based on the PCA model in the present invention. DETAILED DESCRIPTION

[0062] The present invention is further described below in conjunction with some specific embodiments.

[0063] Example 1

[0064] See also Figure 1 A distribution network gale disaster early warning method based on PCA model includes the following steps:

[0065] Step 1: Obtain the main parameters, refined meteorological parameters, and operating status parameters of the distribution network under the gale fault disaster, and establish a distribution network historical gale fault sample library;

[0066] Step 2: Match the entity parameters, refined meteorological parameters and operating status parameters when the fault occurs to form a training sample set;

[0067] Step 3: Use the multiple regression method to fit the power outage duration of the training sample set in step 2 and establish a power outage duration prediction model based on multiple regression;

[0068] Step 4: Based on step 3, the PCA model is used to obtain the principal components of the relevant multivariate parameters in the multivariate regression training process, and the principal component parameters are extracted to establish a distribution network gale disaster power outage duration prediction model based on PCA;

[0069] Step 5: Based on step 3 and combined with the SVM algorithm, the duration of power outages caused by heavy wind disasters in the distribution network is classified by SVM according to the duration of power outages. According to the classification results, the level of heavy wind disasters based on different power outage durations is divided;

[0070] Step 6: In the prediction stage, according to the input distribution network parameters, refined meteorological parameters and operating status parameters, based on the prediction model trained in step 4, the power outage duration of different distribution network lines under strong wind disasters is predicted. Combined with the strong wind disaster level classification in step 5, the early warning of strong wind disasters in the distribution network is finally realized.

[0071] Furthermore, the ontological parameter acquisition method in step 1 is: statistically classify the internal factors that affect the gale disaster in the distribution network, perform ontological parameter analysis, and obtain the ontological parameters that affect the gale disaster in the distribution network.

[0072] Furthermore, the main body parameters include the tower information and line information of the distribution network, and the tower information includes: tower type, number, longitude, latitude, altitude, manufacturer, installation location, geological environment information, horizontal span, vertical span, tower insulator string model, number of strings, number of insulator pieces, and tension tower rotation angle;

[0073] Line information includes: line serial number, voltage level, line number, line name, starting and ending locations, line specifications, line type, number of loops, transmission length, design wind speed, design ice thickness, split number and split gap.

[0074] Furthermore, the meteorological parameter acquisition method in step 1 is: statistically classify the external factors triggered by the strong wind disaster that affects the operation of the distribution network, perform meteorological parameter analysis, and obtain the meteorological parameters that affect the strong wind disaster in the distribution network.

[0075] Furthermore, meteorological parameters include: wind speed, wind direction, line angle, Doppler radar original parameters, whether there is precipitation, and the amount of precipitation.

[0076] Furthermore, the steps for establishing the power outage duration prediction model based on multivariate regression in step 3 are as follows:

[0077] 3.1 Input distribution network gale fault disaster training sample set X:

[0078] X={(X1,y1),(X2,y2),...(X i ,y i ),(X N ,y N )}

[0079] Among them, X i =(x i1 ,x i2 ,...x iM ), X i is the input parameter of the ith gale disaster fault sample, i.e., the distribution network main parameters, meteorological parameters and operation status parameters when the ith gale disaster fault sample occurs, x iM is the Mth parameter in the i-th gale disaster fault sample, y i is the power outage duration when the i-th gale disaster fault sample occurs;

[0080] 3.2 X i Substituting into the multiple regression formula, we get:

[0081] a1x i1 +a2x i2 +...a j x ij +a M x iM +b=y i

[0082] Among them, x ij is the jth parameter in the i-th gale disaster fault sample. Each sample has M parameters, a j is the weight to be trained corresponding to the jth parameter, and b is the constant to be trained;

[0083] Through multivariate regression training, we get the corresponding a j and b.

[0084] Furthermore, the steps for establishing the PCA-based distribution network gale disaster power outage duration prediction model described in step 4 are as follows:

[0085] 4.1 According to the weights (a1, a2, ... a j ...,a M ,b), substitute into the PCA model and get:

[0086]

[0087] in, for The transposed matrix of , B is the PCA principal component model conversion matrix, and we get:

[0088] Where N<M, is the weight of the jth principal component parameter, is the constant obtained through training;

[0089] 4.2 Establish a PCA-based distribution network wind disaster power outage duration prediction model:

[0090]

[0091] y i ' is the predicted duration of power outage.

[0092] Furthermore, in step 5, the steps of classifying the gale disaster levels based on different power outage durations are as follows:

[0093] 5.1 Let φ(X) represent the feature vector after mapping X to high-dimensional space. The hyperplane model of SVM classifier can be expressed as:

[0094] f(X)=w T φ(X)+b

[0095] In the formula, f(X) is the hyperplane model, the training sample set X is the input, w is the weight, and b is the bias vector;

[0096] 5.2 Minimizing the Function

[0097]

[0098] stn i f(X i )≥1(i=1,2,…,m)

[0099] Where n i For X i Type label, X i is the input parameter of the i-th gale disaster fault sample;

[0100] 5.3 Since the formula in step 5.2 is a convex quadratic programming problem, adding the Lagrangian operator to it gives:

[0101]

[0102]

[0103] Introducing the kernel function k(X i ,X j ),satisfy:

[0104] k(X i ,X j )≤φ(X i ),φ(X j )≥φ(Xi ) T φ(X j )

[0105] And the function is obtained as:

[0106]

[0107] Where: X i is the input parameter of the i-th gale disaster fault sample, X j is the input parameter of the jth gale disaster fault sample, n i For X i The type label of k(X i ,X j ) is the kernel function selected by SVM, α i is the Lagrangian operator;

[0108] 5.4 According to the SVM multi-classification strategy, select D main parameters and construct A binary classifier is used, and each classifier is trained for only two parameters. The distribution network gale disaster is divided into D levels according to the power outage duration.

[0109] Furthermore, distribution network gale disasters are divided into four levels according to the duration of power outages, namely: extremely minor power outage accidents (≤0.5h), minor accidents (0.5<t≤2h), general power outage accidents (2h<t≤10h) and major power outage accidents (>10h).

[0110] Furthermore, the early warning steps for the gale disaster in the distribution network described in step 6 are:

[0111] 6.1X c =(x c1 ,x c2 ,...x cN ) is the principal component parameter of the cth case in the prediction stage, x cN is the Nth principal component parameter of the cth case. These principal component parameters include: distribution network parameters, refined meteorological parameters and operation status information parameters. Substituting them into the prediction model trained in step 4, we get:

[0112] Get y c '=t

[0113] Among them, t is the predicted duration;

[0114] 6.2 Based on the prediction of power outage duration in 6.1 and the classification of gale disaster levels in step 5, an early warning of gale disasters in the distribution network can be finally achieved.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Other modifications or equivalent substitutions made to the technical solution of the present invention by ordinary technicians in the field should be included in the scope of the claims of the present invention as long as they do not depart from the spirit and scope of the technical solution of the present invention.

Claims

1. A distribution network gale disaster early warning method based on PCA model, characterized in that: The following steps are involved: Step 1: Obtain the main parameters, refined meteorological parameters, and operating status parameters of the distribution network under the gale fault disaster, and establish a distribution network historical gale fault sample library; Step 2: Match the entity parameters, refined meteorological parameters and operating status parameters when the fault occurs to form a training sample set; Step 3: Use the multiple regression method to fit the power outage duration of the training sample set in step 2 and establish a power outage duration prediction model based on multiple regression; Step 4: Based on step 3, the PCA model is used to obtain the principal components of the relevant multivariate parameters in the multivariate regression training process, and the principal component parameters are extracted to establish a distribution network gale disaster power outage duration prediction model based on PCA; Step 5: Based on step 3 and combined with the SVM algorithm, the duration of power outages caused by heavy wind disasters in the distribution network is classified by SVM according to the duration of power outages. According to the classification results, the level of heavy wind disasters based on different power outage durations is divided; Step 6: In the prediction stage, based on the prediction model trained in step 4 and the classification of gale disasters in step 5, the early warning of gale disasters in the distribution network is finally realized; The steps for establishing the PCA-based distribution network gale disaster power outage duration prediction model described in step 4 are as follows: 4.1 According to the weights (a1, a2, ... a j ...,a M ,b), substitute into the PCA model and get: in, for The transposed matrix of , B is the PCA principal component model conversion matrix, and we get: Where N<M, is the weight of the jth principal component parameter, is the constant obtained through training; 4.2 Establish a PCA-based distribution network wind disaster power outage duration prediction model: y i ' is the predicted duration of power outage.

2. The distribution network gale disaster early warning method based on the PCA model according to claim 1 is characterized in that: The main body parameters described in step 1 include the tower information and line information of the distribution network.

3. The distribution network gale disaster early warning method based on the PCA model according to claim 2 is characterized in that: The tower information includes: tower type, number, longitude, latitude, altitude, manufacturer, installation location, geological environment information, horizontal span, vertical span, tower insulator string model, number of strings, number of insulator pieces, and tension tower rotation angle.

4. The distribution network gale disaster early warning method based on the PCA model according to claim 2 is characterized in that: The line information includes: line serial number, voltage level, line number, line name, starting and ending locations, line specifications, line type, number of loops, transmission length, design wind speed, design ice thickness, split number and split gap.

5. The distribution network gale disaster early warning method based on the PCA model according to claim 1 is characterized in that: The meteorological parameters described in step 1 include: wind speed, wind direction, line angle, Doppler radar original parameters, whether there is precipitation, and precipitation size.

6. The distribution network gale disaster early warning method based on the PCA model according to claim 1 is characterized in that: The steps for establishing the power outage duration prediction model based on multivariate regression described in step 3 are as follows: 3.1 Input distribution network gale fault disaster training sample set X: X={(X1,y1),(X2,y2),...(X i ,y i ),(X N ,y N )} Among them, X i =(x i1 ,x i2 ,...x iM ), X i is the input parameter of the ith gale disaster fault sample, i.e., the distribution network main parameters, meteorological parameters and operation status parameters when the ith gale disaster fault sample occurs, x iM is the Mth parameter in the i-th gale disaster fault sample, y i is the power outage duration when the i-th gale disaster fault sample occurs; 3.2 X i Substituting into the multiple regression formula, we get: a1x i1 +a2x i2 +...a j x ij +a M x iM +b=y i Among them, x ij is the jth parameter in the i-th gale disaster fault sample. Each sample has M parameters, a j is the weight to be trained corresponding to the jth parameter, and b is the constant to be trained; Through multivariate regression training, we get the corresponding a j and b.

7. The distribution network gale disaster early warning method based on the PCA model according to claim 1 is characterized in that: The steps for classifying the gale disaster level based on different power outage durations described in step 5 are as follows: 5.1 Let φ(X) represent the feature vector after mapping X to high-dimensional space. The hyperplane model of SVM classifier can be expressed as: f(X)=w T φ(X)+b In the formula, f(X) is the hyperplane model, the training sample set X is the input, w is the weight, and b is the bias vector; 5.2 Minimizing the Function Where n i For X i Type label, X i is the input parameter of the i-th gale disaster fault sample; 5.3 Since the formula in step 5.2 is a convex quadratic programming problem, adding the Lagrangian operator to it gives: Introducing the kernel function k(X i ,X j ),satisfy: k(X i ,X j )≤φ(X i ),φ(X j )≥φ(X i ) T φ(X j ) And the function is obtained as: Where: X i is the input parameter of the i-th gale disaster fault sample, X j is the input parameter of the jth gale disaster fault sample, n i For X i The type label of k(X i ,X j ) is the kernel function selected by SVM, α i is the Lagrangian operator; 5.4 According to the SVM multi-classification strategy, select D main parameters and construct A binary classifier is used, and each classifier is trained for only two parameters. The distribution network gale disaster is divided into D levels according to the power outage duration.

8. The distribution network gale disaster early warning method based on the PCA model according to claim 7 is characterized in that: The early warning steps for heavy wind disasters in the distribution network described in step 6 are: 6.1X c =(x c1 ,x c2 ,...x cN ) is the principal component parameter of the cth case in the prediction stage, x cN is the Nth principal component parameter of the cth case, and is substituted into the prediction model trained in step 4 to obtain: Get y c '=t Among them, t is the predicted duration; 6.2 Combined with the classification of gale disaster levels in step 5, the early warning of gale disasters in the distribution network can be finally achieved.

Images