CN111539842B - Overhead power transmission line icing prediction method based on meteorological and geographic environments - Google Patents

Abstract

The invention belongs to the technical field of icing prediction, and in particular relates to a method for icing prediction of overhead transmission lines based on weather and geographical environment. The present invention is mainly based on meteorological data and geographic environment data, using machine learning methods to predict and warn the icing situation of overhead transmission lines in the future; train the machine learning model according to the historical icing data of the target line, and solve the category imbalance, Big data training and other issues. The whole process includes data processing, feature extraction and icing prediction model establishment. The present invention realizes icing prediction and early warning of overhead transmission lines through data processing technology, feature extraction technology and classification model establishment on massive data, and combines all historical icing data of transmission lines with the impact factor of icing ( Meteorology, geographical environment, etc.) to realize ice coverage prediction for a period of time in the future.

Description

Icing Prediction Method for Overhead Transmission Lines Based on Meteorological and Geographical Environment

technical field

The invention belongs to the technical field of icing prediction, in particular to an icing prediction method for overhead transmission lines based on meteorological and geographical environments.

Background technique

Icing on transmission lines is one of the most important disasters in power systems. Due to icing, the load on transmission lines increases, causing line disconnection, tower collapse, ice flash tripping, wire galloping, communication interruption, damage to insulators and fittings, etc. ACCIDENT. Icing disasters will seriously threaten the safe and stable operation of the power grid and cause huge economic losses. There are serious icing disasters in many places in our country. Predicting the situation of icing is an important part of power grid operation. Therefore, the prediction of icing on transmission lines is very positive and important to the power grid.

The mechanism of icing on transmission lines is relatively complicated. At present, most studies on icing on transmission lines are based on physical equations, and some studies are conducted through kinetic and thermodynamic experiments. There are also many problems in these methods, such as greater difficulty in application and lower accuracy in application. There are many problems in the application based on the complexity and assumptions of physical equations. The mechanism of icing is complex and difficult to describe through a physical equation, and there are big problems in generality. In addition, general physical equations rely on many empirical coefficients. However, these coefficients are difficult to determine; and physical methods cannot predict and judge future icing conditions.

With the rise of big data and machine learning, there are more and more icing data about transmission lines, so the present invention adopts a machine learning method based on big data to study the icing of transmission lines, and invents a method based on weather and geography Environmental icing prediction methods for overhead transmission lines. The main disadvantages of traditional physical methods are as follows:

1. The physical method is mainly based on the observation of a certain part to give a more reasonable physical equation. When it is used in a large number of applications, the versatility will be a big problem. It must be established for each part (each tower or a section of line) Physical equations, very low applicability.

2. The physical equation itself is constructed with many influencing factors and coefficients, but the determination of the coefficients is generally given based on experience or multiple observations, which makes the determination of the coefficients more complicated and difficult, and the rationality of these coefficients is not easy to scientifically verify. The formation mechanism of icing is extremely complex, and the possible coefficients may be different at different points in time. Therefore, the whole process of constructing physical equations is more difficult.

3. The physical method is mainly based on the current situation to evaluate the current icing state, and cannot predict and evaluate for a period of time in the future.

Contents of the invention

The present invention is mainly based on meteorological data and geographical environment data, and uses machine learning methods to predict and warn the icing situation of overhead transmission lines in the future; trains machine learning models according to historical icing data of target lines, and solves category imbalance, Big data training and other issues. The whole process includes data processing, feature extraction and icing prediction model establishment.

To achieve the above object, the present invention adopts the following technical solutions:

The icing prediction method for overhead transmission lines based on meteorological and geographical environment, as shown in Figure 1, includes the following steps:

S1. Data collection: Collect meteorological data and geographic environment elevation data. Meteorological data include temperature, humidity, rainfall, wind speed, wind direction, wind force level, and date;

S2. Data processing:

Parse the elevation data in hdf format through hdf, and mark the missing value as -1;

Unify the obtained temperature data above 40 degrees Celsius as 40 degrees Celsius;

Set the obtained values of humidity less than 0 to 0, and the values of humidity greater than 100 to 100;

S3. Perform derivative calculations on the collected data to extract more features, specifically:

The slope and aspect are obtained by deriving the elevation data, using the fitting surface method, for the point e, use its 8 nearest data points, upper left, upper, upper right, left, right, lower right, lower, lower right 8 points plus the center point e can form a 3*3 window (matrix) (here is a key problem, because this description is also used in the claims, and the claims cannot refer to the drawings, that is, the attached The picture cannot be used to explain the claim, the claim must only be described in words, so I want to determine whether this window can be described in this way), as shown in Figure 2, the data in the first column are e ₅ , e ₂ , e _6. The data in the second column are e ₁ , e , and e ₃ in sequence, and the data in the third column are e ₈ , e ₄ , and e ₇ in sequence. The calculation formulas for the slope and aspect of the center point e are as follows:

cellsize是像元点的分辨率大小，如果点e周围存钱缺失值，则用邻近的值替代，若都缺少，则点e的坡度和坡向均为0；in,

cellsize is the resolution of the pixel point. If there is a missing value around point e, it will be replaced by the adjacent value. If both are missing, the slope and aspect of point e are both 0;

Perform statistical calculations on the acquired meteorological data, including:

Maximum value v _max =max(v _t ), t=t, t-1, t-2, t-3, t-4,  …

Minimum value v _min =min(v _t ), t=t, t-1, t-2, t-3, t-4,  …

average

standard deviation

Extreme value v _p =v _max -v _min

Among them, v _t refers to meteorological data, and t refers to the latest t time points. Among the obtained meteorological data, the maximum, minimum, mean, standard deviation, and extreme value of the temperature are calculated according to the current value obtained at the current time; the humidity is calculated Calculation of maximum, minimum, average and extreme values; maximum calculation of wind speed; maximum calculation of wind power level; maximum, minimum and extreme calculation of rainfall;

Discretize the obtained wind direction data, because the wind direction data is a character type, convert it into a numerical type, and discretize it into a categorical variable, that is, north, east northeast, northeast, northeast north, east, east southeast, southeast, southeast south, South, Southwest South, Southwest, West Southwest, West, West Northwest, Northwest, Northwest North 16 orientations are converted into integer values from 1 to 16;

S4, judging whether the amount of data obtained in step S3 is greater than 100,000, if so, proceed to step S5, otherwise proceed to step S6;

S5. Based on whether there is an order of magnitude difference of 5 times between the positive samples and negative samples in the icing of overhead transmission lines, the category imbalance is judged. If the difference is 5 times, the category imbalance is determined, and the data is input into the bagging mode model. Otherwise, the machine learning method is used. method to obtain prediction results; the bagging mode model is: the input data set is divided to obtain N subsets, and a training model is constructed for each subset, that is, N training models are constructed, and each subset undergoes a corresponding training model Finally, the results of N training models are obtained, and the bagging algorithm is used to integrate the results of N models to obtain prediction results;

S6. Based on whether there is an order of magnitude difference of 5 times between the positive samples and negative samples in the icing of overhead transmission lines, the category imbalance is judged. If the difference is 5 times, the category imbalance is determined, and the data is input into the weight model. Otherwise, the machine learning method is used. Obtain prediction result; Described weight model is: the weight that gets when objective function training is different, defines objective function as

Where k is the category variable, w is the weight, L is the loss function, y _k is the target variable in the observation sample, f(x _k ) is the output result of the basic model, and the basic model includes logistic regression, support vector machine, decision-making Tree, boosting integrated learning classification method;

The formula for calculating the weight w is:

Among them, n is the number of samples, m _class is the number of samples corresponding to the class, and the value of class is the category of the target variable;

After training according to the obtained weight w, use the trained model to obtain the prediction result.

The beneficial effect of the present invention is that the present invention realizes icing prediction and early warning of overhead transmission lines through data processing technology, feature extraction technology and classification model establishment on massive data, and through all historical icing data of transmission lines, Combining the influencing factors of icing (meteorology, geographical environment, etc.) to realize icing prediction for a period of time in the future; on the basis of big data, combining the formation mechanism of icing, and using machine learning to train and predict the target area. The whole strategy is more flexible, can effectively predict each target area, and is more versatile and convenient in application. It not only overcomes the definite point of physical methods, but also makes the icing prediction of transmission lines more flexible and simple. , easy to apply.

Description of drawings

Fig. 1 is a schematic diagram of the logical structure of the present invention;

Figure 2 is a schematic diagram of a 3*3 data window based on the calculation of slope and aspect based on elevation data;

Fig. 3 is the model schematic diagram that the present invention builds based on bagging mode;

Fig. 4 is the schematic diagram of model training of the present invention;

Fig. 5 is a schematic diagram of AUC and ROC curve indicators of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings.

As shown in Figure 3, the present invention adopts a plurality of sub-model strategies to build a model, and the base classifier of the sub-model selects a simple logistic regression (Logistic) model, for example, the total data set is divided into 41 subsets, and each subset constructs a models, a total of 41 models were constructed. During application, the data will pass through 41 models, and then the results of the 41 models will be output uniformly through voting.

Due to the large amount of data used, batch-by-batch training is adopted during training. And introduce parameter optimization in the training process to output the optimal model. The training process is shown in Figure 4.

After the model is built, the effect of the trained model is evaluated based on the test set. The evaluation index is mainly through AUC and ROC curve. The AUC value of the present invention is 0.9665, and the ROC curve is shown in Figure 5.

Claims (1)

1. The method for predicting icing of overhead transmission lines based on meteorological and geographical environment, is characterized in that, comprises the following steps: S1. Data collection: Collect meteorological data and geographic environment elevation data. Meteorological data include temperature, humidity, rainfall, wind speed, wind direction, wind force level, and date; S2. Data processing: Parse the elevation data in hdf format through hdf, and mark the missing value as -1; Unify the obtained temperature data above 40 degrees Celsius as 40 degrees Celsius; Set the obtained values of humidity less than 0 to 0, and the values of humidity greater than 100 to 100; S3. Perform derivative calculations on the collected data to extract more features, specifically: The slope and aspect are obtained through derivative calculation of elevation data. Using the fitting surface method, for point e, it is surrounded by 8 adjacent data to form a 3*3 window. The data in the first column are e ₅ , e ₂ , e ₆ , the data in the second column are e ₁ , e, e ₃ in sequence, and the data in the third column are e ₈ , e ₄ , e ₇ in sequence, then the calculation formulas for the slope and aspect of the center point e are as follows :

in,

Among them, cellsize is the resolution of the pixel point. If there is a missing value around point e, it will be replaced by an adjacent value. If both are missing, the slope and aspect of point e are both 0; Perform statistical calculations on the acquired meteorological data, including: Maximum value v _max =max(v _t ), t=t, t-1, t-2, t-3, t-4,  … Minimum value v _min =min(v _t ), t=t, t-1, t-2, t-3, t-4,  … average

standard deviation

Extreme value v _p =v _max -v _min Where v _t refers to meteorological data, and t refers to the latest t time points. Among the obtained meteorological data, the current value of temperature and humidity is obtained according to the current time, and the maximum, minimum, mean, standard deviation, and extreme value of the temperature are calculated. ; Calculate the maximum, minimum, average, and extreme values of humidity; calculate the maximum value of wind speed; calculate the maximum value of wind power level; calculate the maximum, minimum, and extreme values of rainfall; Discretize the obtained wind direction data, because the wind direction data is a character type, convert it into a numerical type, and discretize it into a categorical variable, that is, north, east northeast, northeast, northeast north, east, east southeast, southeast, southeast south, South, Southwest South, Southwest, West Southwest, West, West Northwest, Northwest, Northwest North 16 orientations are converted into integer values from 1 to 16; S4, judging whether the amount of data obtained in step S3 is greater than 100,000, if so, proceed to step S5, otherwise proceed to step S6; S5. Based on whether there is an order of magnitude difference of 5 times between the positive samples and negative samples in the icing of overhead transmission lines, the category imbalance is judged. If the difference is 5 times, the category imbalance is determined, and the data is input into the bagging mode model. Otherwise, the machine learning method is used. method to obtain prediction results; the bagging mode model is: the input data set is divided to obtain N subsets, and a training model is constructed for each subset, that is, N training models are constructed, and each subset undergoes a corresponding training model Finally, the results of N training models are obtained, and the bagging algorithm is used to integrate the results of N models to obtain prediction results; S6. Based on whether there is an order of magnitude difference of 5 times between the positive samples and negative samples in the icing of overhead transmission lines, the category imbalance is judged. If the difference is 5 times, the category imbalance is determined, and the data is input into the weight model. Otherwise, the machine learning method is used. Obtain prediction result; Described weight model is: the weight that gets when objective function training is different, defines objective function as

Where k is the category variable, w is the weight, L is the loss function, y _k is the target variable in the observation sample, f(x _k ) is the output result of the basic model, and the basic model includes logistic regression, support vector machine, decision-making Tree, boosting integrated learning classification method; The formula for calculating the weight w is:

Among them, n is the number of samples, m _class is the number of samples corresponding to the class, and the value of class is the category of the target variable; After training according to the obtained weight w, use the trained model to obtain the prediction result.

Images