Short-term icing prediction method based on random forest improved whale optimization algorithm and SVRM

Abstract

The invention discloses an improved whale optimization algorithm and SVRM short-term icing prediction method based on a random forest, and the method comprises the steps: (1) obtaining all micrometeorological data and the equivalent icing thickness thereof, carrying out the mapping along with a time sequence, preprocessing the data, and dividing the data into a training set and a test set; (2) analyzing the influence of each meteorological factor on icing growth based on a random forest algorithm, obtaining an icing influence rate through the random forest algorithm, and taking the characteristic quantity with the high influence rate as an input characteristic quantity; (3) initializing a support vector machine regression model; (4) adding stage position updating in the basic whale optimization algorithm, and adding adaptive weight for improvement; (5) optimizing a kernel parameter and a penalty factor of the support vector machine regression model by using an improved whale algorithm; and (6) reconstructing a support vector machine regression model by using the obtained optimal parameters, training the support vector machine regression model by using the training set, testing the input of the test set, obtaining a final icing prediction value, and performing error calculation on a prediction result.

Description

Technical Field

[0001] The present invention relates to the technical field of ice prediction for power transmission lines. Specifically, the present invention relates to a short-term ice prediction method based on an improved whale optimization algorithm and SVRM under random forest. Further, the present invention relates to a short-term ice thickness prediction method based on RF-GSWOA-SVRM (random forest algorithm-improved whale optimization algorithm-support vector regression machine). Background Art

[0002] The following is a description of the relevant technical background of the present invention, which does not necessarily constitute the prior art of the present invention.

[0003] In recent years, due to the impact of global warming, extreme climates often occur, which can easily cause severe icing of power transmission lines in low temperature and high humidity areas. When transmission lines are covered with ice, they are usually accompanied by a series of line dancing, tripping, line breaking, pole collapse, etc., causing incalculable economic losses to industrial production and people's daily lives. my country is a country with complex terrain and climate conditions. Under different terrain conditions, meteorological factors will vary significantly with the environment, season and vegetation coverage. Micro-meteorological zones are characterized by limitations and complexity. Therefore, the research on the prediction of icing of transmission lines in micro-meteorological zones has important engineering practical value.

[0004] At present, domestic and foreign scholars have conducted extensive research on transmission line icing prediction, mainly focusing on two categories: physical models and statistical models. Icing prediction methods based on physical models are usually complex in calculation, difficult to build, and costly, so they are difficult to be widely used in actual icing prediction. As a result, researchers have started more from statistical regression models to explore the possibility of icing prediction, such as analyzing the impact of meteorological factors on the icing accumulation process, classifying meteorological processes, and further establishing models to predict the growth of icing or optimizing parameters such as weights and thresholds of BP networks through genetic algorithms. However, some of these prediction models may be disturbed when facing noisy data; some models rely on a large amount of historical data support, making it difficult to achieve high-precision predictions when data is scarce; and some models mainly rely on rough weather forecast data to predict ice thickness, which may lead to large prediction errors. Summary of the invention

[0005] The present invention can solve the aforementioned technical problems.

[0006] Specifically, in response to the current status of transmission line icing monitoring, the present invention proposes a random forest algorithm-improved whale optimization algorithm-support vector regression machine (RF-GSWOA-SVRM) method for short-term icing prediction of transmission lines in micro-meteorological zones with strong global search performance and better robustness after studying the application of the SVRM model in the prediction of transmission line icing.

[0007] According to one aspect of the present invention, the present invention provides a short-term ice cover prediction method based on an improved whale optimization algorithm and SVRM under random forest, comprising the following steps:

[0008] (1) Obtain various micrometeorological data and their equivalent ice thickness and map them with time series, preprocess the data, and divide the processed data into training set and test set;

[0009] (2) Based on the random forest algorithm, the impact of various meteorological factors on ice growth is analyzed, and the ice influence rate obtained by the random forest algorithm is taken as the input feature value;

[0010] (3) Initialize the support vector machine regression model SVRM;

[0011] (4) Adding periodic position updates to the basic whale optimization algorithm and adding adaptive weights to improve the optimization capability of the whale optimization algorithm;

[0012] (5) Use the improved whale algorithm GSWOA to optimize the kernel parameters and penalty factors of the support vector machine regression model SVRM;

[0013] (6) The optimal parameters are used to reconstruct the support vector machine regression model SVRM, the training set is used to train the support vector machine regression model SVRM, the test set is input into the support vector machine regression model SVRM for testing, the final ice cover prediction value is obtained, and the relevant error calculation is performed on the prediction result.

[0014] Preferably, in step (1), the micrometeorological data mainly include environmental temperature, air humidity, real-time wind speed, real-time wind direction, atmospheric pressure, solar radiation, rainfall, etc., and the micrometeorological data are matched one by one with equivalent ice thickness data, and finally data samples about time series are obtained, and the first 80% of each combined data set is used as a training set, and the last 20% of the combined data set is used as a test set.

[0015] Preferably, the step (2) comprises: taking the line equivalent ice thickness data as a reference series, taking the micrometeorological data as a comparison series, and normalizing all micrometeorological data.

[0016] The data normalization processing formula is as follows;

[0017]

[0018] In the formula, y is the normalized data; x is the original data, x min and x max They are the minimum and maximum values ​​in the original data, respectively, and the normalized range is [0, 1].

[0019] The random forest algorithm is used to calculate the impact rate, and the impact rates of each comparison sequence are ranked according to their impact rates. The meteorological factors with high impact rates are taken as input.

[0020] Preferably, in step (3), the kernel function k(x i , x j )for:

[0021]

[0022] g is the kernel function parameter, x i and x j is the input sample vector to be predicted.

[0023] Preferably, in step (4), the population size, maximum number of iterations and position dimension of the whale optimization algorithm are set; the current optimal values ​​of all individuals in the population are calculated, and the current optimal whale individual position is obtained and recorded; the current optimal whale individual is searched by automatically adjusting the search radius, and the searched whale individual is used to replace the obtained optimal whale individual;

[0024] Introducing parameter vectors A, C and a and inertia weight ω, the calculation formulas are:

[0025] A=2a·r1-a

[0026] C===2·r2

[0027] a=2-2t / T

[0028]

[0029] Where r1 and r2 are random numbers in the range [0, 1], t represents the current number of iterations, and T represents the maximum number of iterations;

[0030] Determine whether the value of the improved whale algorithm satisfies |A|<1; if so, perform shrinkage and encirclement iterative update on the individual whale position vector, and the update formula is:

[0031] X(t+1)=ω(t)·X rand (t)-A·|C·X rand (t)-X(t)|

[0032] Among them, X(t+1) is the position of the optimal individual at the t+1th iteration in the current space, X(t) * is the position of the current optimal solution, X(t) is the position of the whale in the current iteration, and X rand is the position of an individual whale randomly selected from the current group;

[0033] If not, the individual whale position vector is updated iteratively through random search and predation;

[0034] Introduce parameter p, p is a random number in [0, 1]; determine whether p satisfies p<0.5; if so, introduce the formula:

[0035] X(t+1)=ω(t)·X(t) * -AD;

[0036] Where D = |CX(t) * -X(t)|;

[0037] Otherwise, introduce the formula:

[0038] X(t+1)=ω(t)·X(t) * -D′·e bl ·cos(2π1);

[0039] Where D′=|X(t) * -X(t)|, l is a random number in the range of [-1, 1]. When 1=-1, the individual whale is closest to the food, and when 1=1, the individual whale is farthest from the food.

[0040] Determine whether the maximum number of iterations of the algorithm has been reached. If so, end the process; otherwise, re-enter the loop.

[0041] Specifically, the modified whale optimization algorithm (GSWOA) is used to select the parameters in the SVRM model. The algorithm has faster convergence speed and lower iteration loss, strong optimization ability, and automatically adjusts the search radius to avoid the selection of kernel functions and the setting of penalty factors from falling into the local optimal solution. The algorithm divides the whale population into three types: circling predation, encirclement predation and search predation. In each iteration, the position needs to be updated and the optimal value of fitness in the population is calculated to obtain the position of the global optimal solution until the end of the cycle. The expression of the whale population circling predation is:

[0042]

[0043] In the formula, X(t+1) is the position of the optimal individual at the t+1th iteration in the current space, X(t) *is the position of the current optimal solution, X(t) is the whale position in the current iteration, A and C are intermediate calculation parameters, and

[0044]

[0045] r1 and r2 are random numbers in [0, 1], t represents the current number of iterations, and T represents the maximum number of iterations;

[0046] The expression of whale population encirclement predation is:

[0047]

[0048] Where p is a random number between 0 and 1; D′ is the distance from the whale to its prey, and D′=|X(t) * -X(t)|, l is a random number in the range of (-1, 1).

[0049] The expression for the whale population's search for prey is:

[0050]

[0051] X(t+1)=ω(t)·X(t) * -AD, p<0.5 (6)

[0052] X(t+1)=ω(t)·X(t) * -D′·e bl ·cos(2πl), p≥0.5 (7)

[0053] X(t+1)=ω(t)·X rand (t)-A·|C·X rand (t)-X(t)| (8)

[0054] Where, X rand is the position of an individual whale randomly selected from the current group.

[0055] Preferably, in step (5), when the number of iterations reaches a set maximum number of iterations, an optimal solution is obtained, thereby determining the penalty coefficient c and the kernel function parameter g.

[0056] Specifically, according to step (4), iterative cycles are performed through circling, encircling, searching and preying. When the number of iterations reaches the set maximum number of iterations, the optimal solution is obtained, thereby determining the penalty coefficient c and the kernel function parameter g.

[0057] Preferably, step (6) comprises: inputting the penalty factor c and the optimal solution of the kernel function parameter g output by the improved whale algorithm into the support vector machine regression model SVRM, and inputting the training set data for model training, training the support vector machine regression model SVRM according to the training set, and predicting the ice thickness of the test set, finally obtaining the test set predicted ice thickness data, and calculating the relevant evaluation index of the ice thickness prediction value; the two evaluation indexes used are the mean absolute error MAE and the root mean square error RMSE:

[0058] The formula for calculating the root mean square error is as follows:

[0059]

[0060] The mean absolute error is calculated as follows:

[0061]

[0062] Where: ——predicted value, (in is the predicted value for each moment), y i True value, y i ={y1...y n}(where y1…y n is the true value at each moment), where, when evaluating the model, the smaller the MAE, the higher the overall prediction accuracy of the model; the smaller the RMSE, the higher the prediction accuracy of the model at the extreme point.

[0063] According to the above technical solution of the present invention, the present invention has the following beneficial effects:

[0064] Firstly, the icing process of transmission lines is complex and the data of related meteorological factors are multidimensional, random, large-scale and high-frequency. The RF algorithm is used to extract the characteristic values ​​that affect the icing of transmission lines, so as to reduce the overfitting phenomenon of a certain meteorological factor and the superposition of multiple meteorological factors. Then, the traditional whale algorithm is optimized to obtain the improved whale optimization algorithm (GSWOA) to avoid the kernel function and penalty factor from falling into the local optimal solution. Then, the two parameters of SVRM are optimized by GSWOA algorithm, and the short-term icing prediction model of RF-GSWOA-SVRM is established. Based on this model, the micrometeorological data are associated with the ice thickness, and the correlation between ice thickness, micrometeorology and time is also obtained.

[0065] The short-term ice prediction model based on a small sample data set proposed in the present invention is more suitable for actual engineering life, and has stronger global searchability and better robustness. This model is put into actual engineering, with good results, wide application range, and strong practicality and superiority. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 is a schematic flow chart of a method for predicting short-term icing of a power transmission line according to an embodiment of the present invention;

[0067] Figure 2 A schematic flow chart of an improved whale algorithm optimization process according to an embodiment of the present invention;

[0068] Figure 3 is a schematic diagram of a random forest structure according to an embodiment of the present invention;

[0069] FIG4( a ) is a time series data diagram of temperature of a power transmission line according to an embodiment of the present invention, wherein the horizontal axis is the sample number;

[0070] FIG4( b ) is a time series data diagram of humidity of a power transmission line according to an embodiment of the present invention, wherein the horizontal axis is the sample number;

[0071] FIG4( c ) is a time series data diagram of wind speed of a power transmission line according to an embodiment of the present invention, wherein the horizontal axis is the sample number;

[0072] FIG4( d ) is a time series data diagram of a transmission line with respect to wind direction according to an embodiment of the present invention, wherein the horizontal axis is the sample number;

[0073] FIG4( e ) is a time series data diagram of rainfall for a power transmission line according to an embodiment of the present invention, wherein the horizontal axis is the sample number;

[0074] FIG4( f ) is a time series data diagram of the atmospheric pressure of the power transmission line according to an embodiment of the present invention, wherein the horizontal axis is the sample number;

[0075] FIG4( g ) is a time series data diagram of equivalent ice thickness of a transmission line according to an embodiment of the present invention, wherein the abscissa is the sample number;

[0076] Figure 5 A schematic diagram of screening results of micro-meteorological factors using a random forest algorithm according to an embodiment of the present invention;

[0077] Figure 6 A flowchart of an improved whale optimization algorithm and SVRM prediction model based on random forest according to an embodiment of the present invention;

[0078] Figure 7 is a schematic diagram of errors of a test set according to an embodiment of the present invention;

[0079] Figure 8 Schematic diagram of ice thickness prediction values ​​using different algorithms according to an embodiment of the present invention. DETAILED DESCRIPTION

[0080] Example 1

[0081] Figure 1 is a schematic flow chart of a method for predicting short-term icing of a power transmission line according to an embodiment of the present invention; Figure 2 A schematic flow chart of an improved whale algorithm optimization process according to an embodiment of the present invention; Figure 3 4(a) is a time series data diagram of temperature of a transmission line according to an embodiment of the present invention, wherein the horizontal axis is the sample number; FIG. 4(b) is a time series data diagram of humidity of a transmission line according to an embodiment of the present invention, wherein the horizontal axis is the sample number; FIG. 4(c) is a time series data diagram of wind speed of a transmission line according to an embodiment of the present invention, wherein the horizontal axis is the sample number; FIG. 4(d) is a time series data diagram of wind direction of a transmission line according to an embodiment of the present invention, wherein the horizontal axis is the sample number; FIG. 4(e) is a time series data diagram of rainfall of a transmission line according to an embodiment of the present invention, wherein the horizontal axis is the sample number; FIG. 4(f) is a time series data diagram of atmospheric pressure of a transmission line according to an embodiment of the present invention, wherein the horizontal axis is the sample number; FIG. 4(g) is a time series data diagram of equivalent ice thickness of a transmission line according to an embodiment of the present invention, wherein the horizontal axis is the sample number; Figure 5 A schematic diagram of screening results of micro-meteorological factors using a random forest algorithm according to an embodiment of the present invention; Figure 6 A flowchart of an improved whale optimization algorithm and SVRM prediction model based on random forest according to an embodiment of the present invention; Figure 7 is a schematic diagram of errors of a test set according to an embodiment of the present invention; Figure 8 Schematic diagram of ice thickness prediction values ​​using different algorithms according to an embodiment of the present invention.

[0082] Figure 1 The flowchart of the short-term ice cover prediction method based on the improved whale optimization algorithm and SVRM under random forest is shown, which specifically includes the following steps:

[0083] Step (1): Obtain various micro-meteorological data and their equivalent ice thickness and map them with time series, pre-process the data, and divide the processed data into a training set and a test set;

[0084] Step (2): Based on the random forest algorithm, the impact of the collected meteorological factors on ice growth is analyzed, and the ice impact rate obtained by the random forest algorithm is selected as the input feature quantity;

[0085] Step (3): Initialize the SVRM model (support vector machine regression model);

[0086] Step (4): Add periodic position updates to the basic whale optimization algorithm and add adaptive weights to improve the optimization capability of the whale optimization algorithm;

[0087] Step (5): Use the improved whale algorithm GSWOA to optimize the kernel parameters and penalty factors of the support vector machine regression model SVRM;

[0088] Step (6): Use the obtained optimal parameters to reconstruct the support vector machine regression model SVRM, use the training set to train the support vector machine regression model SVRM, input the test set into the support vector machine regression model SVRM for testing, obtain the final ice cover prediction value, and perform relevant error calculation on the prediction result.

[0089] Specifically, in step (1), first, the micro-meteorological data around the required transmission line and the equivalent ice thickness data of the line are obtained, wherein the micro-meteorological data mainly include, for example, the temperature of the environment, air humidity, real-time wind speed, real-time wind direction, atmospheric pressure, solar radiation, rainfall, etc. The micro-meteorological data are matched one by one with the equivalent ice thickness data, and finally data samples about the time series are obtained, and the first 80% of each combined data set is used as a training set, and the last 20% of the combined data set is used as a test set.

[0090] Furthermore, in step (2), the collected equivalent ice thickness of the transmission line is used as a reference series, and the micro-meteorological data around the line is used as a comparison series.

[0091] In order to avoid a large gap between data values ​​of different eigenvalues, all micrometeorological data are normalized.

[0092] The normalization formula is as follows;

[0093]

[0094] In the formula, y is the normalized data; x is the original data, x min and x max They are the minimum and maximum values ​​in the original data, respectively, and the normalized range is [0, 1].

[0095] Then use the random forest algorithm, the algorithm process is as follows Figure 3As shown, the influence rate is obtained, and each comparison sequence is sorted by influence rate. The meteorological factors with high influence rate are taken as input.

[0096] Furthermore, in step (3), the support vector machine regression model SVRM kernel function k(x i , x j )for:

[0097]

[0098] g is the kernel function parameter, x i and x j is the input sample vector to be predicted.

[0099] Step (4) is as follows:

[0100] The improved whale optimization algorithm (GSWOA) is used to select the parameters in the SVRM model. The algorithm has faster convergence speed and lower iteration loss, strong optimization ability, and can automatically adjust the search radius to avoid the selection of kernel functions and the setting of penalty factors from falling into the local optimal solution. The algorithm divides the whale population into three types: circling predation, encircling predation, and searching predation. In each iteration, the position needs to be updated and the optimal value of fitness in the population needs to be calculated to obtain the position of the global optimal solution until the end of the cycle. The specific operation process of the algorithm is as follows Figure 2 As shown. The expression of whale population surrounding predation is:

[0101]

[0102] In the formula, X(t+1) is the position of the optimal individual at the t+1th iteration in the current space, X(t) * is the position of the current optimal solution, X(t) is the whale position in the current iteration, A and C are intermediate calculation parameters, and

[0103]

[0104] r1 and r2 are random numbers in [0, 1], t represents the current number of iterations, and T represents the maximum number of iterations;

[0105] The expression of whale population encirclement predation is:

[0106]

[0107] Where p is a random number between 0 and 1; D′ is the distance from the whale to its prey, and D′=|X(t) * -X(t)|, l is a random number in the range of (-1, 1).

[0108] The expression for the whale population's search for prey is:

[0109]

[0110] X(t+1)=ω(t)·X rand (t)-A·|C·X rand (t)-X(t)| (8)

[0111] Where, X rand is the position of an individual whale randomly selected from the current group.

[0112] Furthermore, in step (5), iterative cycles are performed through encircling, encircling, searching and preying. When the number of iterations reaches the set maximum number of iterations, the optimal solution is obtained, thereby determining the penalty coefficient c and the kernel function parameter g.

[0113] Furthermore, in step (6), the penalty factor c and the optimal solution of the kernel function parameter g output by the improved whale algorithm are input into the support vector machine regression model SVRM, and the training set data is input for model training. The SVRM model is trained according to the training set, and the ice thickness is predicted for the test set, and finally the predicted ice thickness data of the test set is obtained, as shown in FIG. Figure 6 As shown in the figure, the relevant evaluation indicators of ice thickness prediction value are calculated. Two evaluation indicators are used, namely, mean absolute error (MAE) and root mean square error (RMSE).

[0114] The formula for calculating the root mean square error is as follows:

[0115]

[0116] The mean absolute error is calculated as follows:

[0117]

[0118] Where: — a predicted value, (in is the predicted value at each moment), yi true value, yi={y1...y n} (where y1…yn is the true value at each moment). When evaluating the model, the smaller the MAE, the higher the overall prediction accuracy of the model; the smaller the RMSE, the higher the prediction accuracy of the model at the extreme point.

[0119] Example 2

[0120] In order to verify the effectiveness and feasibility of the short-term icing prediction model based on micrometeorological data provided by the present invention, real-time online monitoring data of a power transmission line in a certain place in Henan Province was selected as a sample for algorithm verification.

[0121] The verification process is as follows:

[0122] Step (1): Obtain various micrometeorological data and their equivalent ice thickness and map them over time series, preprocess the data, and divide the processed data into a training set and a test set.

[0123] Specifically, a 110KV overhead transmission line in Henan Province was selected, and the tension sensor installed on the transmission tower used a formula to convert the equivalent ice thickness of the current line according to the gravity of the line before ice cover and the gravity of the line after ice cover. Micrometeorological sensors installed around the line are used to monitor the current environmental temperature, humidity, wind speed, wind direction, atmospheric pressure, rainfall and other meteorological parameters in real time. The measurement time period of environmental temperature, humidity, wind speed, wind direction, atmospheric pressure, rainfall and other meteorological parameters as well as ice thickness is from 0:00 on January 7, 2024 to 24:00 on January 8, 2024. The sampling time interval is 15 minutes, with a total of 192 groups. The time series measurement results of environmental temperature, humidity, wind speed, wind direction, atmospheric pressure, rainfall and ice thickness are shown as follows. Figure 4(a) to Figure 4(g) As shown;

[0124] Step (2): Based on the random forest algorithm, analyze the impact of the collected meteorological factors on ice growth, and obtain the ice influence rate through the random forest algorithm, and take the one with the highest influence rate as the input feature quantity.

[0125] The collected transmission line equivalent ice thickness is used as the reference series, and the micro-meteorological data around the line is used as the comparison series. In order to avoid the data value gap between different eigenvalues ​​being too large, all micro-meteorological data are normalized.

[0126] The normalization formula is as follows;

[0127]

[0128] In the formula, y is the normalized data; x is the original data, x min and x max They are the minimum and maximum values ​​in the original data, respectively, and the normalized range is [0, 1].

[0129] Then use the random forest algorithm, the algorithm process is as follows Figure 3 As shown in the figure, the influence rate is obtained, and each comparison sequence is sorted by influence rate. The meteorological factors with high influence rate are taken as input. Finally, the relationship between micro-meteorological factors and ice thickness influence rate is obtained, as shown in the figure below: Figure 5 As shown. Figure 5 From the influence rates of various micro-meteorological factors on ice thickness, it can be concluded that temperature and humidity have a strong influence on ice, wind speed and wind direction have a strong influence, and rainfall and atmospheric pressure have a weak influence. Therefore, temperature, humidity, wind speed and wind direction are used as inputs of the ice prediction model, and ice thickness is used as the output of the prediction model.

[0130] Step (3): Initialize the support vector machine regression model SVRM.

[0131] In step (3), the support vector machine regression model SVRM kernel function k(x i , x i )for:

[0132]

[0133] g is the kernel function parameter, x i and x j is the input sample vector to be predicted.

[0134] Step (4): Add periodic position updates to the basic whale optimization algorithm and add adaptive weights to improve the optimization capability of the whale optimization algorithm.

[0135] Specifically, the 192 groups of sample models are divided, the first 80% of each combined data set is used as the training set, the last 20% of the combined data set is used as the test set, the first 153 groups are used as the training set, and the last 39 groups are used as the test set. The model is trained using the training set, and the GSWOA algorithm is used to optimize the penalty factor c and kernel function parameter g of the SVRM model.

[0136] The improved whale optimization algorithm (GSWOA) is used to select the parameters in the SVRM model. The algorithm has faster convergence speed and lower iteration loss, strong optimization ability, and can automatically adjust the search radius to avoid the selection of kernel functions and the setting of penalty factors from falling into the local optimal solution. The algorithm divides the whale population into three types: circling predation, encircling predation, and searching predation. In each iteration, the position needs to be updated and the optimal value of fitness in the population needs to be calculated to obtain the position of the global optimal solution until the end of the cycle. The specific operation process of the algorithm is as follows Figure 2 As shown. The expression of whale population surrounding predation is:

[0137]

[0138] In the formula, X(t+1) is the position of the optimal individual at the t+1th iteration in the current space, X(t) * is the position of the current optimal solution, X(t) is the whale position in the current iteration, A and C are intermediate calculation parameters, and

[0139]

[0140] r1 and r2 are random numbers in [0, 1], t represents the current number of iterations, and T represents the maximum number of iterations;

[0141] The expression of whale population encirclement predation is:

[0142]

[0143] Where p is a random number between 0 and 1; D′ is the distance from the whale to its prey, and D′=|X(t) * -X(t)|, l is a random number in the range of (-1, 1).

[0144] The expression for the whale population's search for prey is:

[0145]

[0146] X(t+1)=ω(t)·X rand (t)-A·|C·X rand (t)-X(t)|

[0147] Where, X rand is the position of an individual whale randomly selected from the current group.

[0148] Step (5): Establish a support vector machine regression prediction model based on the improved whale algorithm optimization, and input iterative cycles through circling, encircling, searching and preying according to step (4). When the number of iterations reaches the set maximum number of iterations, the optimal solution is obtained, thereby determining the penalty coefficient c and the kernel function parameter g.

[0149] Step (6): After optimization, the optimal penalty factor c = 0.3129 and the kernel function parameter g = 7,3612 are output. These two optimized parameters are input into the support vector regression model for ice thickness prediction, and the training set data is input for model training. The SVRM model is trained based on the training set, and the ice thickness of the test set is predicted. Finally, the test set predicted ice thickness data is obtained, as shown in Figure 6 shown.

[0150] According to the optimal parameters calculated by GSWOA algorithm and the model trained by training set samples, 39 groups of test samples were predicted and analyzed. The relative error and absolute error between the test samples and the actual value were as follows: Figure 7 As shown in the figure, it can be seen that the absolute error between the model's predicted results and the actual results is 0.5 cm when the original ice thickness is 16.6 cm, and the maximum relative error is 3%, which is highly accurate.

[0151] In order to more intuitively analyze the advantages of the prediction model of the present invention, the SVRM and RF-WOA-SVRM models are selected for comparison with the model of the present invention. The two evaluation indicators used are the mean absolute error (MAE) and the root mean square error (RMSE). The prediction curves of the test sets of each model are shown in Figure 2. Figure 8The comparison results of the evaluation indicators of the prediction results of each model are shown in Table 1.

[0152] The formula for calculating the root mean square error is as follows:

[0153]

[0154] The mean absolute error is calculated as follows:

[0155]

[0156] Table 1. Comparison of errors of three icing prediction models

[0157]

[0158] As can be seen from Table 1, compared with the SVRM model and the RF-WOA-SVRM model, the mean absolute error MAE of the prediction model RF-GSWOA-SVRM of the present invention is reduced by 16.2% and 4.9% respectively, and the root mean square error RMSE is reduced by 14.4% and 5.8%, and the prediction accuracy is further improved, which further demonstrates the reliability of the prediction model proposed in this paper.

[0159] The present invention adopts RF-GSWOA technology to apply to the support vector machine regression model, which can greatly improve the performance of the prediction model and more accurately predict the ice thickness of the transmission line. It can enable the power department to understand the ice condition of the transmission line more timely and accurately and make corresponding treatment, thereby improving the stability of the power system line operation and significantly improving the economic benefits.

[0160] Although the present invention has been described in detail above with general descriptions and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made based on the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all fall within the scope of protection claimed by the present invention.

Claims

1. A short-term ice prediction method based on improved whale optimization algorithm and SVRM under random forest, characterized in that: The steps include: (1) Obtain various micrometeorological data and their equivalent ice thickness and map them with time series, preprocess the data, and divide the processed data into training set and test set; (2) Based on the random forest algorithm, the impact of various meteorological factors on ice growth is analyzed, and the ice influence rate obtained by the random forest algorithm is taken as the input feature value; (3) Initialize the support vector machine regression model SVRM; (4) Adding periodic position updates to the basic whale optimization algorithm and adding adaptive weights to improve the optimization capability of the whale optimization algorithm; (5) Use the improved whale algorithm GSWOA to optimize the kernel parameters and penalty factors of the support vector machine regression model SVRM; (6) The optimal parameters are used to reconstruct the support vector machine regression model SVRM, the training set is used to train the support vector machine regression model SVRM, the test set is input into the support vector machine regression model SVRM for testing, the final ice cover prediction value is obtained, and the relevant error calculation is performed on the prediction result.

2. The method according to claim 1, characterized in that In step (1), the micro-meteorological data mainly include environmental temperature, air humidity, real-time wind speed, real-time wind direction, atmospheric pressure, solar radiation, and rainfall. The micro-meteorological data are matched one by one with equivalent ice thickness data, and finally data samples about the time series are obtained. The first 80% of each combined data set is used as a training set, and the last 20% of the combined data set is used as a test set.

3. The method according to claim 2, characterized in that The step (2) comprises: The line equivalent ice thickness data is used as a reference series, and the micro-meteorological data is used as a comparison series. All micrometeorological data are normalized. The data normalization formula is as follows: y is the normalized data; x is the original data, x min and x max They are the minimum and maximum values ​​in the original data, respectively, and the normalized range is [0,1]; The random forest algorithm is used to calculate the impact rate, and each comparison sequence is sorted by impact rate, and the meteorological factors with high impact rate are taken as input.

4. The method according to claim 3, characterized in that In the step (3), The kernel function k(x i ,x j )for: g is the kernel function parameter, x i and x j is the input sample vector to be predicted.

5. The method according to claim 4, characterized in that In the step (4), the population size, maximum number of iterations and position dimension of the whale optimization algorithm are set; the current optimal values ​​of all individuals in the population are calculated, and the current optimal whale individual position is obtained and recorded; the current optimal whale individual is searched by automatically adjusting the search radius, and the searched whale individual is used to replace the obtained optimal whale individual; Introducing parameter vectors A, C and a and inertia weight ω, the calculation formulas are: A=2a·r1-a C＝2·r2 a=2-2t / T Among them, r1 and r2 are random numbers in [0,1], t represents the current number of iterations, and T represents the maximum number of iterations; Determine whether the value of the improved whale algorithm satisfies |A|<1; if so, perform shrinkage and encirclement iterative update on the individual whale position vector, and the update formula is: X(t+1)=ω(t)·X rand (t)-A·|C·X rand (t)-X(t)| Among them, X(t+1) is the position of the optimal individual at the t+1th iteration in the current space, X(t) * is the position of the current optimal solution, X(t) is the position of the whale in the current iteration, and X rand is the position of an individual whale randomly selected from the current group; If not, the individual whale position vector is updated iteratively through random search and predation; Introduce parameter p, p is a random number in [0,1]; determine whether p satisfies p<0.5; if so, introduce the formula: X(t+1)=ω(t)·X(t) * -AD; Where D = |CX(t) * -X(t)|; Otherwise, introduce the formula: X(t+1)=ω(t)·X(t) * -D′·e bl ·cos(2πl); Where D′=|X(t) * -X(t)|, l is a random number in the range [-1,1]. When l = -1, the individual whale is closest to the food, and when l = 1, the individual whale is farthest from the food. Determine whether the maximum number of iterations of the algorithm has been reached. If so, end the process; otherwise, re-enter the loop.

6. The method according to claim 5, characterized in that In step (5), when the number of iterations reaches the set maximum number of iterations, the optimal solution is obtained, thereby determining the penalty coefficient c and the kernel function parameter g.

7. The method according to claim 6, characterized in that The step (6) comprises: inputting the penalty factor c and the optimal solution of the kernel function parameter g output by the improved whale algorithm into the support vector machine regression model SVRM, and inputting the training set data for model training, training the support vector machine regression model SVRM according to the training set, and predicting the ice thickness of the test set, finally obtaining the test set predicted ice thickness data, and calculating the relevant evaluation index of the ice thickness prediction value; the two evaluation indexes used are the mean absolute error MAE and the root mean square error RMSE: The calculation formula of mean square error is: The calculation formula for the mean absolute error is: Where: ——predicted value, (in is the predicted value for each moment), y i True value, y i ={y1...y n }(where y1...y n is the true value at each moment), where, when evaluating the model, the smaller the MAE, the higher the overall prediction accuracy of the model; the smaller the RMSE, the higher the prediction accuracy of the model at the extreme point.

Images