Fusion algorithm-based power transmission line icing prediction method and system

Abstract

The invention relates to the technical field of power transmission line icing monitoring, in particular to a power transmission line icing prediction method and system based on a fusion algorithm. According to the power transmission line icing prediction method based on the fusion algorithm, a visual image of a power transmission line is collected, the fuzzy degree of the visual image is judged, and a fuzzy degree value Yi is obtained according to the fuzzy degree; obtaining the reflectivity Z of the rain measuring radar in the area where the power transmission line is located; and then substituting the fuzzy degree value Yi and the rain measuring radar reflectivity Z into an icing degree prediction model, calculating to obtain an icing evaluation value E, and outputting an icing degree grade corresponding to the icing evaluation value E. According to the method, the icing degree grade is predicted in combination with the fuzzy category of the shot image and the rainfall, and the icing degree grade is matched with the actually measured icing thickness. Therefore, real-time monitoring of icing of the power transmission line is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of power transmission line icing monitoring, and in particular to a power transmission line icing prediction method and system based on a fusion algorithm. Background Art

[0002] Icing disasters are one of the main causes of transmission line failures. At present, the main methods for monitoring icing include weighing method, conductor inclination method, distributed optical fiber method and image monitoring method. Among them, the weighing method is widely used because of its scientific and reliable calculation model, but the device requires the power supply of the transmission line to be shut down for installation, and the tension metering device needs to be calibrated regularly, which is difficult to meet the actual needs of the power grid; the accuracy of the conductor inclination method mainly depends on the angle sensor and its installation position, and its application effect in actual transmission lines is not good; the distributed optical fiber method can realize real-time and arbitrary position monitoring of the transmission line, but there is a problem of low monitoring accuracy; for the image monitoring method, existing studies have used data classification methods and deep learning algorithms to process visual images based on massive historical image monitoring data to realize the identification of transmission line icing types. The image detection method has realized the identification of icing types to a certain extent, but it has not comprehensively considered the accuracy of the data source and its feasibility in actual projects. Restricted by factors such as icing on transmission lines and environmental climate, icing images have poor shooting quality and low utilization, which makes it difficult to meet the needs of icing type identification. In summary, the traditional ice monitoring method cannot meet the actual operation requirements of the power grid in terms of economy and data reliability. Summary of the invention

[0003] In order to overcome the defect of the above-mentioned prior art that there is a lack of reliable transmission line icing monitoring method, the present invention proposes a transmission line icing prediction method based on a fusion algorithm, which calculates the transmission line icing assessment value by combining the image blur degree category and rainfall, thereby judging the degree of icing of the transmission line and realizing real-time monitoring of the icing situation of the transmission line.

[0004] The present invention proposes a method for predicting ice coating on a power transmission line based on a fusion algorithm, comprising the following steps:

[0005] S1. Construct an icing degree prediction model to calculate the icing evaluation value E:

[0006] E=aR+bY i

[0007] Among them, a is the weight value reflecting the influence of rainfall on ice cover, b is the weight value reflecting the influence of image blur on ice cover; Y i Indicates the blur level value corresponding to the image blur category; Z is the reflectivity of the rain radar; R is the rainfall;

[0008] S2. Collect the visual image of the transmission line, judge the blurring degree of the visual image, and obtain the blurring degree value Y according to the blurring degree. i ; And obtain the rainfall radar reflectivity Z of the area where the transmission line is located.

[0009] S3. Substitute the blurring degree value Y i and the rainfall radar reflectivity Z into the icing degree prediction model, calculate the icing evaluation value E, and output the icing degree level corresponding to the icing evaluation value E.

[0010] Preferably, the icing degree level corresponding to the icing evaluation value E is obtained according to the following strategy:

[0011] If 0 < E ≤ 2b, it is judged that there is no icing in the monitoring area.

[0012] If 2b < E ≤ 5a + 2b and R ≤ 5, it is judged that the monitoring area has light to moderate icing.

[0013] If E > 5a + 2b and R > 5, it is judged that the monitoring area has severe icing.

[0014] R is the rainfall.

[0015] Preferably, the acquisition of the rainfall R includes the following steps:

[0016] St0. Combine historical data to obtain the first prediction model RF, the second prediction model predict and the rainfall prediction weighted model: The first prediction model RF is used to predict the rainfall sequence according to the rainfall radar reflectivity sequence in a specified time period; The second prediction model predict is used to predict the current rainfall according to the known rainfall sequence and the current rainfall radar reflectivity; The rainfall prediction weighted model performs weighted calculation on multiple rainfalls at consecutive times to generate the rainfall prediction value at the next time.

[0017] St1. Obtain multiple rainfall radar reflectivities at specified consecutive times, and perform window sliding sampling on the rainfall radar reflectivities to obtain multiple rainfall radar reflectivity sequences.

[0018] St2. Substitute the rainfall radar reflectivity sequence into the first prediction model RF to obtain the first rainfall prediction value sequence corresponding to each rainfall radar reflectivity sequence.

[0019] St3. Take the average of the first rainfall prediction values corresponding to the same time in different first rainfall prediction value sequences as the preliminary rainfall prediction value at that time; Perform sliding window sampling on the preliminary rainfall prediction values at specified consecutive times to obtain the preliminary rainfall prediction value sequence.

[0020] St4, substituting the rainfall preliminary prediction value sequence and the rain measuring radar reflectivity at the next time point of the sequence into the second prediction model, and obtaining the rainfall prediction value at the next time point of the rainfall preliminary prediction value sequence as the rainfall value to be corrected at the next time point;

[0021] St5, traverse the sequence of preliminary rainfall prediction values, obtain the rainfall values ​​to be corrected at each time in the specified continuous time, and substitute them into the rainfall prediction weighted model to calculate the rainfall prediction value at the next time in the specified continuous time.

[0022] Preferably, the rainfall prediction weighted model is obtained in the following manner:

[0023] First, construct a rainfall prediction weighted model:

[0024]

[0025] Among them, β 0 is the intercept term; ∈ is the set error; β 1 is the weight of the first time in the past, β 2 is the weight of the second time before, β w is the weight of the w-th time in the past; represents the rainfall at time t2-1, represents the rainfall at time t2-2, represents the rainfall at time t2-w; y' t2 Represents the predicted rainfall value at time t2.

[0026] Combine historical data to get a data set In the dataset The genetic algorithm is used to optimize the parameter β of the weighted model. 0 , β 1 , β 2 , …, β w , until the rainfall prediction weighted model converges.

[0027] Preferably, the first prediction model adopts a random forest model.

[0028] Preferably, in S1, firstly, a data set {rainfall, image blur category; ice coverage level} is obtained, then an ice coverage prediction model is trained on the data set, and a and b weight values ​​are iterated using a particle swarm optimization algorithm.

[0029] Preferably, the image blur categories include: clear, light to medium blur and heavy blur, let i∈{0, 1, 2}, where the blur value Y corresponding to “clear” is 0 = 0, the blur value Y corresponding to "mild to medium blur"1 = 1, the blur value Y corresponding to "heavy blur" 2 =2.

[0030] The present invention proposes a system for predicting ice coverage of power transmission lines using the fusion algorithm-based method, comprising: a data acquisition device, an image fuzzy judgment module, a rainfall prediction module and a fusion calculation module;

[0031] The data acquisition device communicates with external equipment to obtain the reflectivity of the rain radar in the monitoring area and take images;

[0032] The image blur judgment module is connected to the data acquisition module, and the image blur judgment module identifies the blur category of the captured image;

[0033] The rainfall prediction module is connected to the data acquisition module, the rainfall prediction module obtains and stores the reflectivity of the rain measuring radar, and the rainfall prediction module predicts the rainfall according to the reflectivity of the rain measuring radar in continuous time;

[0034] The fusion calculation module is connected to the image blur judgment module and the rainfall prediction module respectively. The fusion calculation module obtains the blur degree value corresponding to the blur category of the captured image, and substitutes the blur degree value and the predicted rainfall into the icing degree prediction model to calculate the icing evaluation value E; the fusion calculation module judges the icing degree level according to the icing evaluation value E and outputs it.

[0035] The present invention proposes a transmission line icing prediction device based on a fusion algorithm, comprising a memory and a processor, wherein a computer program is stored in the memory, and the processor is connected to the memory, and the processor is used to execute the computer program to implement the transmission line icing prediction method based on the fusion algorithm.

[0036] The present invention provides a storage medium storing a computer program, which is used to implement the transmission line icing prediction method based on a fusion algorithm when executed.

[0037] The advantages of the present invention are:

[0038] (1) The present invention proposes a transmission line icing prediction method based on a fusion algorithm, which combines the fuzzy category of the captured image and the rainfall to predict the icing degree, and the icing thickness is consistent with the measured icing thickness. It can be seen that the present invention realizes real-time monitoring of icing on transmission lines.

[0039] (2) The present invention predicts the precipitation at the next time based on the reflectivity of the rain measuring radar obtained in the most recent period, and then predicts the degree of ice coverage by combining the precipitation and the image blur category. This realizes the prediction of the degree of ice coverage at the next time by combining the blur category of the current captured image and the reflectivity of the rain measuring radar obtained in the most recent period. The degree of ice coverage can be predicted in advance, so that emergency protection measures can be initiated in time to ensure the safe operation of the transmission line.

[0040] (3) Compared with the fixed single classical ZR relationship, the present invention combines the first prediction model, the second prediction model and the rainfall prediction weighted model to predict rainfall based on the rain radar reflectivity sequence. It has greater flexibility and accuracy, can be applied to the ZR nonlinear relationship in different scenarios, and can better fit the ZR relationship in different terrain areas, geographical conditions and raindrop spectrum distribution. This proves that the rainfall prediction method in the present invention has a strong ability to deal with complex nonlinear relationships.

[0041] (4) The present invention realizes regression analysis in different time periods by introducing a dynamic rolling window mechanism, which can effectively capture the changes between the reflectivity Z of the rain radar and the precipitation intensity R, so that the rainfall prediction method can adapt to the complex regional environment and time-varying raindrop spectrum distribution, and improve the accuracy and stability of rainfall prediction by constructing multiple decision trees.

[0042] (5) The present invention adopts a multi-stage combined rainfall prediction method, which has good flexibility and scalability, can process large-scale data sets, and adjust the window size and model parameters as needed to adapt to precipitation estimation tasks under different meteorological conditions. At the same time, it can better overcome the errors caused by rainfall radar observation noise, complex and changeable precipitation types, etc., and has good robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flow chart of a method for constructing a rainfall prediction model;

[0044] FIG2 is a flow chart of a rainfall prediction method;

[0045] FIG2( a ) is the rainfall distribution obtained by the 16-point prediction in Example 1;

[0046] Figure 2(b) shows the actual rainfall distribution at 16 points in Example 1;

[0047] FIG3( a ) is the rainfall distribution obtained by the 17-point prediction in Example 1;

[0048] Figure 3(b) shows the actual rainfall distribution at 17 points in Example 1;

[0049] Figure 4is the fitting relationship between the rain radar reflectivity and precipitation at 17 points in Example 1;

[0050] Figure 5 It is a flow chart of a transmission line icing prediction method based on fusion algorithm;

[0051] Figure 6 The flowchart of the training method for the image fuzzy classification model;

[0052] Figure 7 This is the structure diagram of the image fuzzy classification model;

[0053] Figure 8 To improve the structure diagram of the Mamba network;

[0054] Figure 9(a) is a clear image;

[0055] Figure 9(b) shows a light to moderate blurred image;

[0056] Figure 9(c) is a heavily blurred image;

[0057] Fig.10 Comparison of loss values ​​of the models in the embodiments;

[0058] Fig.11 The accuracy comparison of the models in the embodiments;

[0059] Fig.12 2 is a comparison of the recall rates of the models in the embodiments. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] Example 1

[0062] Reference Figure 1 , this embodiment proposes a method for constructing a rainfall prediction model, which specifically requires constructing a first prediction model, a second prediction model and a rainfall prediction weighted model respectively.

[0063] The first prediction model RF is used in the known data set Get it through training;

[0064] The first prediction model RF is:

[0065]

[0066]

[0067]

[0068] T is the sliding window length; is the rain radar reflectivity sequence, x t-T+1 represents the reflectivity of the rain radar at time t-T+1, x t-T+2 represents the reflectivity of the rain radar at time t-T+2, x t represents the reflectivity of the rain radar at time t;

[0069] is the rainfall series, y t-T+1 represents the rainfall at time t-T+1 in the output data of the first prediction model, y t-T+2 represents the rainfall at time t-T+2 in the output data of the first prediction model, y t Represents the rainfall at time t in the output data of the first prediction model.

[0070] When the first prediction model RF is applied, y t-T+1 ,y t-T+2 , …, y t is the rainfall prediction value obtained by the first prediction model RF based on the reflectivity of the rain radar at the corresponding time; during the training process of the first prediction model RF, y t-T+1 ,y t-T+2 , …, y t is the actual value of rainfall from the dataset;

[0071] The first prediction model RF can specifically adopt a random forest model; Random Forest is a machine learning algorithm based on decision trees, mainly used for classification and regression tasks. It consists of multiple decision trees, each of which is trained independently. The core idea of ​​random forest is to improve the accuracy and stability of the model by combining the prediction results of multiple decision trees. The core concepts of random forest are "randomness" and "forest". Randomness means that when training decision trees, there is a certain degree of randomness in the construction process of each tree, such as the random selection of features and the random selection of training samples. Forest refers to a collection of multiple decision trees. Therefore, the mathematical principle of random forest mainly involves how to improve the prediction performance of the model through randomness and the structure of the forest.

[0072] Dataset The construction method is as follows: first, obtain the site precipitation data and rain radar monitoring data, clean the data, and then obtain the mapping relationship between rainfall and rain radar reflectivity based on spatial matching. Specifically, first extract the minute-level precipitation data of all stations in the specified area in real time, and accumulate and preprocess the minute-level precipitation data; data preprocessing includes cleaning of obviously abnormal data; jigsaw networking of rain radar monitoring data, converting rain radar data into three-dimensional grid data, processing the rain radar reflectivity of the quality-controlled volume scan data, generating three-dimensional grid data of a single station, and then jigsaw processing of multiple rain radars to obtain the reflectivity grid data of the rain radar network, that is, the mapping relationship between the reflectivity of each rain radar and space; finally, spatially match the rain radar reflectivity data with the site precipitation data to form a sample data pair of rainfall and rain radar reflectivity. For machine learning algorithms to perform model training.

[0073] The second prediction model predicts in the data set The second prediction model predict is:

[0074]

[0075] in, represents the rainfall at time t1+1 output by the second prediction model predict; t1-T+1 represents the rainfall at time t1-T+1 in the input data of the second prediction model, y t1-T+2 represents the rainfall at time t1-T+2 in the input data of the second prediction model, y t1 Represents the rainfall at time t1 in the input data of the second prediction model.

[0076] The second prediction model predict collects a neural network model, specifically a convolutional neural network CNN; during the training process of the second prediction model, y t1-T+1 ,y t1-T+2 ,……,y t1 Actual rainfall values ​​are used.

[0077] The rainfall prediction weighted model is:

[0078]

[0079] Precipitation prediction weighted model in the dataset The genetic algorithm is used to optimize the parameter β of the weighted model. 0 , β 1 , β 2 , …, β w ;

[0080] Among them, β 0 is the intercept term; ∈ is the set error; β 1 is the weight of the first time in the past, β 2 is the weight of the second time period, β w is the weight of the w-th time in the past; w is the set step size, which is specifically set to 5;

[0081] represents the rainfall at time t2-1 in the input data of the rainfall prediction weighted model, represents the rainfall at time t2-2 in the input data of the rainfall prediction weighted model, Represents the rainfall at time t2-w in the input data of the rainfall prediction weighted model;

[0082] During the training of the rainfall prediction weighted model, The actual values ​​of rainfall at time t2-1, t2-2, and t2-w are used respectively.

[0083] The rainfall prediction model proposed in this embodiment is essentially a combination model composed of the first prediction model, the second prediction model and the rainfall prediction weighted model.

[0084] In order to verify the effectiveness of the rainfall prediction method using the combined model, this implementation takes the precipitation event in a certain place from 16:00 to 17:00 on July 5, 2022 as an example to evaluate the performance of the combined model in real-time quantitative precipitation estimation.

[0085] 2 , in this embodiment, a combined model consisting of a first prediction model, a second prediction model and a rainfall prediction weighted model is firstly trained on a known data set, and then rainfall prediction is performed on the data set.

[0086] The rainfall y at time t0 t0 The prediction method includes the following steps St1-St5:

[0087] St1, first obtain the reflectivity of the rain radar at the most recent time t0-wT, t0-w-T+1, t0-w-T+2, ..., t0-1, respectively, denoted as: x t0-w-T 、x t0-w-T+1 、x t0-w-T+2 , …, x t0-1 ; Use a sliding window of length T to analyze the data sequence x t0-w-T 、x t0-w-T+1 、x t0-w-T+2 , …, x t0-1 Perform window sampling to obtain the data segment {(x t'-T+1 ,x t'-T+2 ,…,x t');t0-w-1≤t'≤t0-2}

[0088] {(x t'-T+1 ,x t'-T+2 ,…,x t' );t0-w-1≤t'≤t0-2}

[0089] ={(x t0-w-T ,x t0-w-T+1 ,…,x t0-w-1 );(x t0-w-T+1 ,x t0-w-T+2 ,…,x t0-w );…;(x t0-2-T+1 ,x t0-2-T+2 ,…,x t0-2 )}

[0090] In this embodiment, the sampling time interval of the rain radar data is set to 6 minutes, that is, the interval between two adjacent times in this embodiment is 6 minutes; T is set to 5, and w is set to 5.

[0091] St2, the data fragment x t'-T+1 ,x t'-T+2 ,…,x t' Substituting into the first prediction model RF, we can get the first predicted value y of rainfall at time t'-T+1, t'-T+2, ..., t' t'-T+1 ,y t'-T+2 ,…,y t' ;

[0092] St3, calculate the mean of the first predicted values ​​of rainfall corresponding to different data segments at the same time as the preliminary predicted value of rainfall at that time; calculate the preliminary predicted values ​​of rainfall at time t0-wT, t0-w-T+1, t0-w-T+2, ..., t0-1 Perform window sliding slicing with a window length of T to obtain a sequence of preliminary rainfall prediction values

[0093] St4. Sequence the initial rainfall forecast value and the rain radar reflectivity x t'+1 Substitute into the second prediction model prdict to obtain the rainfall forecast value at time t'+1

[0094]

[0095] St5. Execute St4 on t0-w-1≤t'≤t0-2, and get Will Substituting into the rainfall prediction weighted model, we get:

[0096]

[0097] y' t0 is the final rainfall forecast value at time t0.

[0098] In this embodiment, the cumulative result of the rainfall prediction value at 16:00 on the same day and the distribution of the measured cumulative precipitation are shown in Figures 2(a) and 2(b); the cumulative result of the rainfall prediction value at 17:00 on the same day and the distribution of the measured cumulative precipitation are shown in Figures 3(a) and 3(b);

[0099] It can be seen that the intensity distribution form of the rainfall prediction value of this embodiment is very similar to the actual measured data of the station, with high consistency, and the rainfall prediction accuracy is as high as 93% or more, which can accurately realize the quantitative estimation of precipitation. According to statistics, the cumulative results of the rainfall prediction values ​​of the station in this embodiment are very close to the actual monitored precipitation of the station, and good accuracy is achieved in most areas, with an overall error RMSE of only 1.71 mm.

[0100] In this embodiment, the mapping relationship between the rainfall prediction value and the rain radar reflectivity is evaluated in combination with the data at 17:00 on the day. The mapping relationship between the rainfall prediction value and the rain radar reflectivity obtained by the above steps St1-St5 is as follows: Figure 4 As shown, it can be seen that the mapping relationship between the predicted value and the reflectivity of the rain measuring radar is highly consistent with the mapping relationship between the actual rainfall value and the reflectivity of the rain measuring radar, and both are in line with the ZR curve.

[0101] It can be seen that the above rainfall prediction methods have good performance and robustness, and can achieve good results in different terrain areas, geographical conditions and raindrop spectrum distribution. Figure 4 It can be seen that the predicted value of the 1-hour precipitation distribution is basically consistent with the actual value result, and can better reflect the details of heavy precipitation, making up for the problem of insufficient spatial distribution of rain gauges.

[0102] Example 2

[0103] In this embodiment, taking the ice coverage prediction of a certain transmission line as an example, Figure 5 The transmission line icing prediction method based on the fusion algorithm shown in the figure is verified.

[0104] In this embodiment, firstly, the transmission line image is collected by the online monitoring device installed on the transmission line tower, and the image blur category is marked, and the blur degree value Y corresponding to each image blur category is assigned. i .

[0105] In this embodiment, the image blur categories include: clear, light to medium blur and heavy blur. Let i∈{0, 1, 2}, where the blur value Y corresponding to “clear” is 0= 0, the blur value Y corresponding to "mild to medium blur" 1 = 1, the blur value Y corresponding to "heavy blur" 2 =2.

[0106] In this embodiment, the rainfall prediction method provided in Embodiment 1 is used to obtain the rainfall prediction value R at the next time.

[0107] In this embodiment, the blur level value Y corresponding to the currently collected transmission line image is i Substitute the predicted rainfall value R at the next time into the following formula to calculate the icing assessment value E of the transmission line at the next time:

[0108] E=0.5R+0.5Y i

[0109] In this embodiment, according to the following judgment conditions, the predicted value of the ice coverage level is obtained, and the measured ice thickness, image blur level, measured rainfall and ice coverage level predicted by the model of some monitoring results are statistically analyzed, as shown in Table 1.

[0110]

[0111] In this embodiment, a=b=0.5 is set, then 5a=2.5, 2b=1, 5a+2b=3.5.

[0112] Table 1 Evaluation results of ice coverage level

[0113]

[0114] In the line names in Table 1, 220kV, 800kV and 500kV represent voltage levels, SQ, SN, JT, AQ, LS, GJ and SA represent line abbreviations, # represents the line tower number, “±” represents a DC transmission line, and no “±” represents an AC transmission line.

[0115] It can be seen from Table 1 that ice thickness of 14 mm or less is predicted as light to moderate ice, and ice thickness of 25 mm or more is predicted as heavy ice. It can be seen that the predicted value of ice degree is consistent with the trend of the measured result, which proves the reliability of the ice prediction method provided by the present invention.

[0116] In this embodiment, the weights a and b are simply set to 0.5. In specific implementation, the particle swarm optimization algorithm can be used to iterate the weight values ​​of a and b on the data set {image blur category, rainfall; ice coverage level} to further improve the calculation accuracy of the ice coverage prediction model, thereby improving the accuracy of ice coverage prediction.

[0117] The rain measuring radar reflectivity in this embodiment adopts the rain measuring radar reflectivity provided by the weather forecast system.

[0118] In this embodiment, the icing level at the next time is predicted by combining the blur category of the current captured image and the reflectivity of the rain radar obtained in the recent period, and it is consistent with the ice thickness at the next time. It can be seen that the present invention can predict the icing level in advance, so as to start the protective measures in time and ensure the working safety of the power transmission line.

[0119] In this embodiment, the image blur category is marked using a three-round three-person voting mechanism. In specific implementation, the image blur category can be marked using any existing method.

[0120] Example 3

[0121] This embodiment provides an image fuzzy classification method, which classifies the fuzziness of a visualized image based on an image fuzzy classification model.

[0122] The image blur classification model performs machine learning on the labeled dataset {visualized image, blur category}, and is used to determine the image blur category after the model training converges.

[0123] The image blur level can be specifically divided into: clear, light to medium blur and heavy blur.

[0124] Reference Figure 6 In this embodiment, the training of the image fuzzy classification model includes the following steps:

[0125] S1. Build a basic model and annotate a dataset. The annotated dataset is used to store visualization images annotated with blur classification labels, i.e., blur categories. The blur classification labels in the annotated dataset are annotated using a three-round three-person voting mechanism.

[0126] S2, extract training samples from the labeled data set and let the basic model learn the training samples;

[0127] S3, extract verification samples from the annotated data set and input them into the basic model, obtain the predicted labels output by the basic model, and determine whether the basic model has converged; if not, execute step S4; if yes, fix the basic model as the image fuzzy classification model;

[0128] S4. Calculate the model loss by combining the predicted label and the true label on the verification sample, update the parameters of the base model through back propagation of the model loss, and then return to step S2.

[0129] In the specific implementation, in S4, the model loss adopts the cross entropy loss.

[0130] Reference Figure 7The basic model in this implementation manner includes: a Stem-P network and four Mamba-E networks connected in sequence; the input of the Stem-P network is the input of the basic model, the output of the Stem-P network is connected to the input of the first Mamba-E network, the output of the first Mamba-E network is connected to the input of the second Mamba-E network, the output of the second Mamba-E network is connected to the input of the third Mamba-E network, the output of the third Mamba-E network is connected to the input of the fourth Mamba-E network, and the output of the fourth Mamba-E network is the output of the basic model.

[0131] The initial input image size of the base model is H×W×C, and each subsequent Mamba-E network successively reduces the image size to H / 4×W / 4×C, H / 8×W / 8×2C, H / 16×W / 16×4C, and the last Mamba-E network outputs an image size of H / 32×W / 32×8C.

[0132] During the image processing process, the basic model reduces the image size and increases the channel dimension through progressive downsampling to achieve efficient image feature extraction, thereby generating the final blur degree classification label.

[0133] The Stem-P network consists of sequentially connected convolutional layers, pooling layers, linear layers, and nonlinear feature networks. The Stem-P network divides the input image into patches by downsampling the input feature map using an improved "three layers one network" operation, thereby generating a feature map with a spatial dimension of H / 4×W / 4. The Mamba-E network is used to create hierarchical representations with resolutions of H / 4×W / 4, H / 8×W / 8, H / 16×W / 16, and H / 32×W / 32, and performs downsampling operations through patch merging.

[0134] Reference Figure 8 , the Mamba-E network includes a first layer normalized network, a first multi-layer perceptron, a second layer normalized network, a Mamba network, a first dimension superposition unit, a third layer normalized network, a second multi-layer perceptron, and a second dimension superposition unit connected in sequence; wherein the input of the first layer normalized network is the input of the Mamba-E network, and the output of the second dimension superposition unit is the output of the Mamba-E network; the input of the Mamba-E network is also connected to the input of the first dimension superposition unit, and the output of the first dimension superposition unit is also connected to the input of the second dimension superposition unit;

[0135] That is, the input image features of the Mamba-E network are processed by the first normalization network, the first multi-layer perceptron, the second normalization network and the Mamba network in sequence, the output data of the Mamba network is dimensionally superimposed with the input image data of the Mamba-E network and then processed by the third normalization network and the second multi-layer perceptron again, and then the output data output by the first dimension superposition unit and the output data of the second multi-layer perceptron are dimensionally superimposed by the second dimension superposition unit as the output data of the Mamba-E network; which is equivalent to the output data of the Mamba-E network being the dimensionally superimposed result of the input data of the Mamba-E network, the output data of the Mamba network and the output data of the second multi-layer perceptron.

[0136] In this implementation, the Mamba network in the Mamba-E network adopts a bidirectional sequence, wherein feature x is convolved in a horizontally leftward and vertically upward direction during forward processing, and feature x is convolved in a vertically downward and horizontally rightward direction during backward processing.

[0137] Specifically, the network structure of the Mamba network includes: a normalization layer, an activation function, a forward convolution layer, a forward SSM, a backward convolution layer, a backward SSM, a third dimension stacking unit, and a fourth dimension stacking unit. The forward convolution layer and the forward SSM are connected to form a forward branch, and the backward convolution layer and the backward SSM are connected to form a backward branch.

[0138] The input of the Mamba network is converted into features z and x after being processed by the normalization layer. On the forward branch, feature x is convolved horizontally to the left and vertically upward by the forward convolution layer, and the forward SSM processes the convolution data output by the forward convolution layer as the output of the forward branch. On the backward branch, feature x is convolved vertically downward and horizontally to the right by the backward convolution layer, and the backward SSM processes the convolution data output by the backward convolution layer as the output of the backward branch.

[0139] The output of the backward branch is convolved with the data of the feature z after being activated by the activation function and then input into the third dimension superposition unit. The output of the forward branch is convolved with the data of the feature z after being activated by the activation function and then input into the third dimension superposition unit. The third dimension superposition unit superimposes the two input data. The output of the third dimension superposition unit and the output of the normalization layer are processed by the fourth dimension superposition unit as the output of the Mamba network.

[0140] The above-mentioned image fuzzy classification model is verified in combination with specific embodiments below.

[0141] In this embodiment, more than 70,000 sets of RGB three-channel images collected by visual online monitoring devices were collected from a certain local power transmission panoramic intelligent management and control platform. The collected images were cleaned and reviewed by a three-person voting mechanism to obtain the original image marked with a blur classification label. The original image resolution was compressed to an image sample of 640×640; then the compressed image sample was enhanced, including random cropping, random rotation, horizontal flipping, vertical stretching, changing brightness, etc., to increase the number of samples. Random cropping is to randomly select a method to crop the image from the center, upper left, lower left, upper right, and lower right, and the cropping ratio is set to 0.8; random rotation is to rotate the image at a random angle of -30 degrees to 30 degrees; changing brightness is to perform a random brightness transformation of 0.5 to 1.5 on the image. In this embodiment, the image categories include: clear, light to medium blur, and heavy blur, with specific reference to Figures 9(a), 9(b), and 9(c).

[0142] In this embodiment, all image samples are divided into a training set and a validation set according to a ratio of 8:2; in this implementation, the image fuzzy classification model provided by the present invention is combined with multiple comparison models in the above steps SA1-SA4 to be trained on the training set, and the performance of the converged model is tested on the validation set.

[0143] The comparison models include: YOLOv8x network model using ResNet-18 network, YOLOv8x network model using ResNet-101 network and Mamba-1 network model. The Mamba-1 network model is composed of Stem network and four traditional Mamba networks connected in sequence. The Stem network preprocesses the input image samples, and the Mamba network downsamples the input image features in turn to extract image features and output blur classification labels.

[0144] In this implementation, during the training process of the YOLOv8x network model YOLOv8x (ResNet-18) using the ResNet-18 network, the YOLOv8x network model YOLOv8x (ResNet-101) using the ResNet-101 network, the Mamba-1 network model, and the model of the present invention on the training set, the loss value, precision, and recall rate of the model are as follows: Fig.10 , Fig.11 , Fig.12 As shown, it can be seen that in the training set, the performance indicators of the model of the present invention are higher than those of the comparison model.

[0145] In this embodiment, when the four models are verified on the verification set, the performance of each fuzzy category is shown in Table 2 below.

[0146] Table 2 Comparison of model test results

[0147]

[0148]

[0149] It can be seen from Table 2 above that compared with the YOLOv8x (ResNet-18) model, the model of the present invention has obvious improvements in accuracy, recall rate and F1 value; compared with the YOLOv8x (ResNet-101) model, great progress has been made in accuracy, recall rate and F1 value; it can be seen that the model of the present invention has achieved a certain significant improvement in model performance compared with the YOLOv8x framework.

[0150] From Table 2, it can be seen that the model of the present invention has made progress in the reasoning time compared with the three comparison models, especially compared with the YOLOv8x (ResNet-101) model, the reasoning time has been greatly shortened, as shown in Table 3 for details.

[0151] Table 3 Comparison of inference time of three models

[0152]

[0153] It can be seen from Table 3 that in terms of reasoning time, the model of the present invention saves 0.08 seconds on clear images, 0.139 seconds on lightly and moderately blurred images, and 0.071 seconds on heavily blurred images compared to the YOLOv8x (ResNet-101) model, and the calculation speed is increased by up to 51.29%; the model of the present invention saves 0.013 seconds on clear images, 0.012 seconds on lightly and moderately blurred images, and 0.002 seconds on heavily blurred images compared to the Mamba-1 model, and the calculation speed is increased by up to 8.33%.

[0154] It can be seen that the model of the present invention has either achieved significant improvement in model performance indicators or greatly improved computing speed compared to different YOLOV8 frameworks; compared to the Mamba-1 model, the model of the present invention has not only improved model performance, but also further improved the model's computational reasoning efficiency.

[0155] Of course, it is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, but also includes the same or similar structures that can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-limiting, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any reference numerals in the claims should not be regarded as limiting the claims involved.

[0156] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

[0157] The techniques, shapes, and structural parts not described in detail in the present invention are all well-known techniques.

Claims

1. A transmission line icing prediction method based on a fusion algorithm, characterized in that: It includes the following steps: S1. Construct an icing degree prediction model to calculate the icing evaluation value E: E=aR+bY i Among them, a is the weight value reflecting the influence of rainfall on ice cover, b is the weight value reflecting the influence of image blur on ice cover; Y i Indicates the blur level value corresponding to the image blur category; Z is the reflectivity of the rain radar; R is the rainfall; S2. Collect the visual image of the transmission line, determine the blurriness of the visual image, and obtain the blurriness value Y according to the blurriness i ; and obtain the rain radar reflectivity Z in the area where the transmission line is located; S3, the blur value Y i The reflectivity Z of the rainfall radar is substituted into the icing degree prediction model to calculate the icing assessment value E, and the icing degree level corresponding to the icing assessment value E is output.

2. The transmission line icing prediction method based on fusion algorithm according to claim 1, characterized in that: The icing degree level corresponding to the icing evaluation value E is obtained according to the following strategy: If 0 < E ≤ 2b, it is determined that the monitored area has no icing; If 2b < E ≤ 5a + 2b and R ≤ 5, it is determined that the monitored area has light to moderate icing; If E > 5a + 2b and R > 5, it is determined that the monitored area has severe icing; R is the rainfall.

3. The transmission line icing prediction method based on fusion algorithm according to claim 2, characterized in that: The acquisition of the rainfall R includes the following steps: St0. Combine historical data to obtain the first prediction model RF, the second prediction model predict, and the rainfall prediction weighted model: The first prediction model RF is used to predict the rainfall sequence based on the rain gauge radar reflectivity sequence in a specified time period; The second prediction model predict is used to predict the current rainfall based on the known rainfall sequence and the current rain gauge radar reflectivity; The rainfall prediction weighted model performs weighted calculation on multiple rainfalls at consecutive times to generate the rainfall prediction value at the next time; St1. Obtain multiple rain gauge radar reflectivities at specified consecutive times, and perform window sliding sampling on the rain gauge radar reflectivities to obtain multiple rain gauge radar reflectivity sequences; St2. Substitute the rain gauge radar reflectivity sequence into the first prediction model RF to obtain the first rainfall prediction value sequence corresponding to each rain gauge radar reflectivity sequence; St3. Take the average of the first rainfall prediction values corresponding to the same time in different first rainfall prediction value sequences as the preliminary rainfall prediction value at this time; Perform sliding window sampling on the preliminary rainfall prediction values at specified consecutive times to obtain the preliminary rainfall prediction value sequence; St4. Substitute the preliminary rainfall prediction value sequence and the rain gauge radar reflectivity at the next time point of this sequence into the second prediction model to obtain the rainfall prediction value at the next time point of the preliminary rainfall prediction value sequence as the rainfall value to be corrected at the next time point; St5. Traverse the preliminary rainfall prediction value sequence to obtain the rainfall values to be corrected at each time in the specified consecutive times, and substitute them into the rainfall prediction weighted model to calculate the rainfall prediction value at the next time in the specified consecutive times.

4. The transmission line icing prediction method based on fusion algorithm according to claim 3, characterized in that: The acquisition method of the rainfall prediction weighted model is: First, construct the rainfall prediction weighted model: Among them, β0 is the intercept term; ∈ is the set error; β1 is the weight of the first time in the past, β2 is the weight of the second time in the past, and β w is the weight of the w-th time in the past; represents the rainfall at time t2-1, represents the rainfall at time t2-2, represents the rainfall at time t2-w; y' t2 Represents the predicted rainfall value at time t2. Combine historical data to get a data set In the dataset The genetic algorithm is used to optimize the parameters β0, β1, β2, ..., β w , until the rainfall prediction weighted model converges.

5. The method for predicting icing of power transmission lines based on a fusion algorithm according to claim 2, characterized in that: The first prediction model adopts a random forest model.

6. The method for predicting icing of power transmission lines based on a fusion algorithm according to claim 2, characterized in that: In S1, first obtain the data set {rainfall, image blur category; icing degree level}, then train the icing degree prediction model on the data set, and use the particle swarm optimization algorithm to iterate the a and b weight values.

7. The method for predicting icing of power transmission lines based on a fusion algorithm according to claim 1, characterized in that: The image blur categories include: clear, light to moderate blur, and severe blur. Let i ∈ {0, 1, 2}, where the blur degree value Y0 corresponding to "clear" is 0, the blur degree value Y1 corresponding to "light to moderate blur" is 1, and the blur degree value Y2 corresponding to "severe blur" is 2.

8. A system using the power transmission line icing prediction method based on the fusion algorithm as claimed in claim 3, characterized in that: It includes: A data acquisition device, an image blur judgment module, a rainfall prediction module, and a fusion calculation module; The data acquisition device communicates with external devices and is used to obtain the rain gauge radar reflectivity and captured images of the monitored area; The image blur judgment module is connected to the data acquisition module, and the image blur judgment module identifies the blur category of the captured image; The rainfall prediction module is connected to the data acquisition module, the rainfall prediction module obtains and stores the reflectivity of the rain measuring radar, and the rainfall prediction module predicts the rainfall according to the reflectivity of the rain measuring radar in continuous time; The fusion calculation module is connected to the image blur judgment module and the rainfall prediction module respectively. The fusion calculation module obtains the blur degree value corresponding to the blur category of the captured image, and substitutes the blur degree value and the predicted rainfall into the ice coverage prediction model to calculate the ice coverage evaluation value E; The fusion calculation module determines the ice degree level according to the ice assessment value E and outputs it.

9. A transmission line icing prediction device based on a fusion algorithm, characterized in that: It includes a memory and a processor, the memory stores a computer program, the processor is connected to the memory, and the processor is used to execute the computer program to implement the transmission line icing prediction method based on the fusion algorithm as described in any one of claims 1-8.

10. A storage medium, characterized in that: A computer program is stored, and when the computer program is executed, it is used to implement the transmission line icing prediction method based on the fusion algorithm as described in any one of claims 1 to 8.

Images