Second-order integrated icing prediction method and device based on multi-source icing monitoring data

Abstract

The application provides a second-order integrated ice-coating prediction method and device based on multi-source ice-coating monitoring data, which comprises the following steps: monitoring time series data of a power transmission line under a specified microtopography meteorological scenario, and extracting features of the time series data of the power transmission line; establishing a prediction model of ice-coating sequence data by using the features extracted from the time series data; acquiring ice-coating monitoring image data of the power transmission line, and extracting features of the ice-coating image of the power transmission line; establishing a prediction model of the ice-coating image data by using the features extracted from the ice-coating image of the power transmission line; constructing a multi-source data integrated prediction model according to the prediction model of the ice-coating sequence data and the prediction model of the ice-coating image data and by using a simple average integration method; and using the multi-source data integrated prediction model to acquire a final ice-coating thickness prediction result. The application can accurately predict the line ice-coating condition under a micro-meteorological environment, and ensure the safe and stable operation of the line.

Description

Technical Field

[0001] This invention relates to the field of transmission line icing monitoring technology, and more specifically, to a second-order integrated icing prediction method and device based on multi-source icing monitoring data. Background Technology

[0002] Icing on transmission lines is a significant factor affecting the safe and stable operation of the power grid. Icing can lead to excessive loads on towers, increased conductor sag, and conductor / ground wire jumping during ice removal. In recent years, large-scale icing accidents have occurred frequently across the country, and the accidents involving power outages and tower collapses caused by icing on transmission lines are becoming increasingly serious.

[0003] Micrometeorological conditions for transmission line icing refer to localized areas within a larger region experiencing more severe icing conditions than the larger area due to unique variations in topography, location, slope aspect, temperature, and humidity. This type of icing under micrometeorological conditions is characterized by its small scale and high degree of concealment, making it difficult for transmission line designers, operators, and maintenance personnel to implement anti-icing measures. Therefore, it is necessary to predict icing conditions in such cases to take preventative measures and avoid accidents.

[0004] Current transmission line icing detection models mainly consist of traditional mathematical statistical models and image detection techniques. Relying solely on mathematical statistical models or image detection techniques will lead to problems such as poor robustness and overfitting, making it difficult to accurately predict the icing situation of transmission lines under special conditions such as micrometeorology. Therefore, there is an urgent need for an icing prediction model with high accuracy. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art.

[0006] Therefore, the first aspect of the present invention provides a second-order integrated icing prediction method based on multi-source icing monitoring data.

[0007] A second aspect of the present invention provides a computer device.

[0008] This invention provides a second-order integrated icing prediction method based on multi-source icing monitoring data, comprising: The monitoring time series data of transmission lines under specified micro-topographic meteorological scenarios were analyzed, and features were extracted from the time series data of transmission lines. A predictive model for icing sequence data is established using features extracted from time series data. Acquire icing monitoring image data of transmission lines and extract features from the icing images of transmission lines; A predictive model for icing image data is established using features extracted from icing images of transmission lines. Based on the prediction models of the icing sequence data and the icing image data, and using a simple average ensemble method, a multi-source data ensemble prediction model is constructed, which is then used to obtain the final icing thickness prediction result.

[0009] The second-order integrated icing prediction method based on multi-source icing monitoring data according to the above-described technical solution of the present invention may also have the following additional technical features: In the above technical solution, the time series data includes sensor data collected in time order. The sensor data includes at least one of line tension, fiber optic status, micro-meteorology, and microwave status. The micro-meteorology includes wind speed, temperature, and precipitation.

[0010] In the above technical solution, an autoencoder algorithm is used to extract features from the time-series data of transmission lines, including: A segment of micrometeorological data was selected as the sample set, and the data was divided into datasets by grouping data within a preset time interval. , ... ,in , , R represents the feature dimension matrix, p represents the number of samples, and q represents the dimension of the micro-meteorological features. This represents a dimension of the line's tensile strength characteristics; The raw data acquired by the sensor is reconstructed using an encoder-decoder approach. The encoding process of the raw data X acquired by the sensor from the input layer to the hidden layer includes:

[0011] Where h represents a latent variable, and For the encoder weight bias, Here, X represents the activation function of the encoding layer, where X represents either data sample X1 or X2 in dataset D. The data decoding process from the hidden layer to the output layer includes:

[0012] in, Data representing reconstruction, and For the weight bias of the decoder, The activation function for the decoding layer; The objective function of the autoencoder algorithm is:

[0013] Here, dist represents the vector distance calculation function.

[0014] In the above technical solution, the step of establishing a prediction model for icing sequence data using features extracted from time series data includes: By using a long short-term memory network and training and debugging the model with training and test sets, a predictive model for icing sequence data is established.

[0015] in, This represents the time-series characteristics of icing. This represents the mapping relationship between icing thickness and icing time-series characteristic data. This represents the output of the time series prediction model.

[0016] In the above technical solution, the step of establishing a prediction model for icing image data based on features extracted from transmission line icing images includes: Features of icing images were extracted using the ResNet-18 model, and then the network model was trained to build a prediction model for icing image data.

[0017] in, This represents the feature data of the icing image. This indicates the mapping relationship between ice thickness and ice image data. This represents the output of the image data prediction model.

[0018] In the above technical solution, the step of constructing a multi-source data ensemble prediction model based on the prediction model of the icing sequence data and the prediction model of the icing image data, and using a simple averaging ensemble method, includes: Weights are assigned based on the similarity between micro-meteorological data and micro-meteorological data of typical micro-topography in historical statistics; A simple average ensemble method is used to fuse the prediction models for the icing sequence data and the icing image data; Multiple multi-source data integration prediction models were established under various micro-topographic meteorological scenarios, and the final ice thickness prediction result was obtained by integrating the information from multiple models through adaptive weighted integration.

[0019] In the above technical solution, the weight allocation based on the similarity between micro-meteorological data and micro-meteorological data of typical micro-topography in historical statistics includes: The micro-meteorological data of the target route are clustered with typical micro-meteorological data in historical statistics to obtain multiple clusters, each cluster corresponding to a cluster center; Map the target route data to these cluster centers and calculate the Euclidean distance between them; The Euclidean distance from the target route data to a cluster center is used to characterize the similarity between the target route data and historical micro-meteorological data. Weights are then assigned to each micro-topographic meteorological scenario model based on this similarity. The calculation method for assigning weights to each micro-topographic meteorological scenario model includes:

[0020]

[0021] Where di represents the similarity between the target route data and the i-th micro-topography meteorological scenario model, i.e., the Euclidean distance; n represents the number of micro-topography meteorological scenario models; This represents the sum of similarity scores between each of the n micro-topographic meteorological scenario models and the target route data; The coordinates of the i-th cluster center are represented by x and y; the coordinates of the target line data are represented by x and y.

[0022] In the above technical solution, the step of fusing the prediction model of the icing sequence data and the prediction model of the icing image data using a simple average ensemble method includes: A performance-weighted approach was adopted to evaluate the weights of the prediction models for icing sequence data and icing image data by using the model's performance on the validation set. The prediction model is run on both the icing sequence data and the icing image data. The prediction results are recorded on the same validation dataset. Then, the prediction error of each model is calculated. The prediction error is used as the model performance score. The weights a and b of the two prediction models are obtained based on the performance score. The prediction results for the scene are obtained based on the weights of the two prediction models:

[0023] Where 'a' represents the ensemble weights of the prediction model for the icing sequence data, and 'b' represents the ensemble weights of the prediction model for the icing image data. Representing a scene The prediction results of the prediction model for the icing sequence data are as follows. Representing a scene The prediction results of the prediction model for the icing image data are as follows: Representing a scene The final prediction result.

[0024] In the above technical solution, the step of establishing multiple multi-source data integration and prediction models under multiple micro-topographic meteorological scenarios, and obtaining the final icing thickness prediction result by adaptively weighted integration of information from multiple models, includes:

[0025] in, This indicates the final predicted icing thickness.

[0026] The present invention also provides a computer device, the computer device including a processor and a memory, the memory storing a computer program, the computer program being loaded and executed by the processor to implement the second-order integrated icing prediction method based on multi-source icing monitoring data as described in any of the above technical solutions.

[0027] In summary, due to the adoption of the above-mentioned technical features, the beneficial effects of the present invention are: This invention integrates multi-source data in the icing prediction model, effectively consolidating different data sources and providing comprehensive icing prediction information. This method is capable of processing different types of data, improving the diversity and adaptability of data processing. This invention employs a multi-stage integration and adaptive weighting method to enhance the adaptability and accuracy of icing prediction, thereby maintaining the high stability and reliability of the power system.

[0028] Additional aspects and advantages of the invention will become apparent in the following description or may be learned by practice of the invention. Attached Figure Description

[0029] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a flowchart of a second-order integrated icing prediction method based on multi-source icing monitoring data according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the principle of data reconstruction of raw data collected by sensors in a second-order integrated icing prediction method based on multi-source icing monitoring data according to an embodiment of the present invention. Figure 3 This is a framework diagram of the second-order integrated icing prediction method based on multi-source icing monitoring data according to an embodiment of the present invention, which outputs the final icing prediction result. Detailed Implementation

[0030] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0031] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0032] The following reference Figures 1 to 3 This invention describes a second-order integrated icing prediction method based on multi-source icing monitoring data, provided by some embodiments of the present invention.

[0033] Some embodiments of this application provide a second-order integrated icing prediction method based on multi-source icing monitoring data.

[0034] like Figure 1 As shown, the first embodiment of the present invention proposes a second-order integrated icing prediction method based on multi-source icing monitoring data, including steps S1-S5.

[0035] S1. Monitor the time series data of the transmission line under the specified micro-topographic meteorological scenario, and extract features from the time series data of the transmission line; specifically, the specified micro-topographic meteorological scenario refers to a scenario where the topographic conditions and some climate conditions are known, and the time series data includes sensor data collected in time order, and the sensor data includes line tension, fiber optic status, micro-meteorological conditions and microwave status, and the micro-meteorological conditions include wind speed, temperature and precipitation.

[0036] S2. Build a prediction model for icing sequence data using features extracted from time series data.

[0037] S3. Acquire icing monitoring image data of transmission lines and extract features from the icing images of transmission lines; S4. Establish a prediction model for icing image data using features extracted from icing images of transmission lines; S5. Based on the prediction model of the icing sequence data and the prediction model of the icing image data, and using a simple average ensemble method, construct a multi-source data ensemble prediction model, and use the multi-source data ensemble prediction model to obtain the final icing thickness prediction result.

[0038] It should be noted that the steps S1-S5 above are not in a fixed order and can be flexibly adjusted according to the actual operation.

[0039] Specifically, the aforementioned icing prediction method includes prediction and early warning for transmission lines under different mountainous terrains, and can be used for prediction and early warning of transmission lines under various mountainous terrains. Ice thickness is the detection target. It integrates various data sources such as tension, icing, rainfall, wind speed, and images, and provides an icing thickness prediction model through second-order ensemble. Multiple algorithms are applied, including a time-series data model based on autoencoders and long short-term memory networks, and an image data model based on ResNet-18. These models play a key role in the integration of multi-source data and adaptive weighted ensemble to achieve accurate prediction of transmission line icing thickness. Integrating multiple data sources and models into a comprehensive prediction system improves the maintainability and operational safety of transmission lines in mountainous areas.

[0040] In some embodiments, step S1 includes: A segment of micrometeorological data was selected as the sample set, and the data was divided into datasets by grouping data within a preset time interval. , ... ,in , , R represents the feature dimension matrix, p represents the number of samples, and q represents the dimension of the micro-meteorological features. This represents the dimension of the line tension characteristics; the preset interval length can be one day, but in other embodiments it can also be other lengths, such as half a day, two days, etc.

[0041] like Figure 2 As shown, an encoding-decoding method is used to reconstruct the raw data acquired by the sensor. The encoding process of the raw data X acquired by the sensor from the input layer to the hidden layer includes:

[0042] Where h represents a latent variable, and For the encoder weight bias, Here, X represents the activation function of the encoding layer, where X represents either data sample X1 or X2 in dataset D. The data decoding process from the hidden layer to the output layer includes:

[0043] in, Data representing reconstruction, and For the weight bias of the decoder, The activation function for the decoding layer; The objective function of the autoencoder algorithm is:

[0044] Where dist represents the vector distance calculation function, specifically: .

[0045] In some embodiments, step S2 includes: Based on the extracted time-series data features, and using a Long Short-Term Memory (LSTM) network, a model is trained and debugged using training and testing sets to establish a prediction model for icing sequence data.

[0046] in, This represents the time-series characteristics of icing. This represents the mapping relationship between icing thickness and icing time-series characteristic data. This represents the output of the time series prediction model.

[0047] In some embodiments, step S4 includes: Features of icing images were extracted using the ResNet-18 model, and then the network model was trained to build a prediction model for icing image data.

[0048] in, This represents the feature data of the icing image. This indicates the mapping relationship between ice thickness and ice image data. This represents the output of the image data prediction model.

[0049] It is understood that the above-mentioned feature extraction methods and separate prediction model establishment methods are not limitations of this disclosure. Those skilled in the art can also use other model establishment methods, as long as they can obtain prediction models for icing sequence data and prediction models for icing image data.

[0050] In some embodiments, step S5 includes S51-S53.

[0051] S51. Based on the similarity between micro-meteorological data and micro-meteorological data of typical micro-topography in historical statistics, weights are assigned.

[0052] Specifically, step S51 includes: The micro-meteorological data of the target route are clustered with typical micro-meteorological data in historical statistics to obtain multiple clusters, each cluster corresponding to a cluster center.

[0053] The target route data is mapped to these cluster centers, and the Euclidean distance between them is calculated to assess the similarity between the target route data and specific micro-topographic historical data. The smaller the value, the higher the similarity between the target route and the scenario model, and the higher the weight needs to be assigned accordingly.

[0054] The Euclidean distance from the target route data to a cluster center is used to characterize the similarity between the target route data and historical micro-meteorological data. Weights are then assigned to each micro-topographic meteorological scenario model based on this similarity. To ensure a total weight of 1, the Euclidean distance is converted into a similarity score, and then the similarity is normalized to a unit length. For example, the calculation method for assigning weights to each micro-topographic meteorological scenario model includes:

[0055]

[0056] Where di represents the similarity between the target route data and the i-th micro-topography meteorological scenario model, i.e., the Euclidean distance; n represents the number of micro-topography meteorological scenario models; This represents the sum of similarity scores between each of the n micro-topographic meteorological scenario models and the target route data; The coordinates of the i-th cluster center are represented by x and y; the coordinates of the target line data are represented by x and y.

[0057] S52. The prediction model of the icing sequence data and the prediction model of the icing image data are fused using the simple average ensemble (first-order ensemble) method.

[0058] Specifically, step S52 includes: In order to reasonably integrate the two models and obtain a more accurate prediction result, a performance weighting method is adopted to evaluate the weights of the prediction model for icing sequence data and the prediction model for icing image data by using the model's performance on the validation set. The prediction model is run on both the icing sequence data and the icing image data. The prediction results are recorded on the same validation dataset. Then, the prediction error of each model is calculated. The prediction error is used as the model performance score. The weights a and b of the two prediction models are obtained based on the performance score. The prediction results for the scene are obtained based on the weights of the two prediction models:

[0059] Where 'a' represents the ensemble weights of the prediction model for the icing sequence data, and 'b' represents the ensemble weights of the prediction model for the icing image data. Representing a scene The prediction results of the prediction model for the icing sequence data are as follows. Representing a scene The prediction results of the prediction model for the icing image data are as follows: Representing a scene The final prediction result.

[0060] S53, such as Figure 3 As shown, multiple multi-source data integration prediction models are established under multiple micro-topographic meteorological scenarios, and the final ice thickness prediction result is obtained by integrating the information of multiple models through adaptive weighted integration.

[0061] Specifically, step S53 includes:

[0062] in, This indicates the final predicted icing thickness.

[0063] Multiple prediction models were established based on different micro-topography and meteorological scenarios. Then, an adaptive weighted ensemble method was used to calculate the Euclidean distance between the target route data and the scene center, subsequently determining the scene similarity, automatically assigning weights to each scenario, and finally calculating the prediction results for the route data under each scenario. This allows for accurate prediction of icing thickness on transmission lines. Integrating multiple models can reduce prediction biases that may arise from relying on a single model. Different models may exhibit advantages in different scenarios; integrated prediction leverages these advantages, combining information from multiple models to obtain more accurate results. This improves reliability and adaptability under various conditions.

[0064] A second embodiment of the present invention provides a computer device, the computer device including a processor and a memory, the memory storing a computer program, the computer program being loaded and executed by the processor to implement the second-order integrated icing prediction method based on multi-source icing monitoring data as described in any of the above embodiments.

[0065] In this specification, the illustrative expressions of the terms used do not necessarily refer to the same embodiments or examples. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0066] Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this invention shall be included within the scope of protection of this invention.

Claims

1. A second-order integrated icing prediction method based on multi-source icing monitoring data, characterized in that, include: The monitoring time series data of transmission lines under specified micro-topographic meteorological scenarios were analyzed, and features were extracted from the time series data of transmission lines. A predictive model for icing sequence data is established using features extracted from time series data. Acquire icing monitoring image data of transmission lines and extract features from the icing images of transmission lines; A predictive model for icing image data is established using features extracted from icing images of transmission lines. Based on the prediction models of the icing sequence data and the icing image data, and using a simple average ensemble method, a multi-source data ensemble prediction model is constructed, which is then used to obtain the final icing thickness prediction result. The step of constructing a multi-source data ensemble prediction model based on the prediction model of the icing sequence data and the prediction model of the icing image data, using a simple averaging ensemble method, includes: Weights are assigned based on the similarity between micro-meteorological data and micro-meteorological data of typical micro-topography in historical statistics; A simple average ensemble method is used to fuse the prediction models for the icing sequence data and the icing image data; Multiple multi-source data integration and prediction models were established under various micro-topographic meteorological scenarios, and the information from multiple models was integrated through adaptive weighted integration to obtain the final ice thickness prediction result. The method of fusing the prediction models for the icing sequence data and the icing image data using a simple average ensemble includes: A performance-weighted approach was adopted to evaluate the weights of the prediction models for icing sequence data and icing image data by using the model's performance on the validation set. The prediction model is run on both the icing sequence data and the icing image data. The prediction results are recorded on the same validation dataset. Then, the prediction error of each model is calculated. The prediction error is used as the model performance score. The weights a and b of the two prediction models are obtained based on the performance score. The prediction results for the scene are obtained based on the weights of the two prediction models: Where 'a' represents the ensemble weights of the prediction model for the icing sequence data, and 'b' represents the ensemble weights of the prediction model for the icing image data. Representing a scene The prediction results of the prediction model for the icing sequence data are as follows. Representing a scene The prediction results of the prediction model for the icing image data are as follows: Representing a scene The final prediction result.

2. The second-order integrated icing prediction method based on multi-source icing monitoring data according to claim 1, characterized in that, The time-series data includes sensor data collected in chronological order, which includes line tension and micro-meteorological data, including wind speed, temperature, and precipitation.

3. The second-order integrated icing prediction method based on multi-source icing monitoring data according to claim 2, characterized in that, Feature extraction of time-series data from transmission lines is performed using an autoencoder algorithm, including: A segment of micrometeorological data was selected as the sample set, and the data was divided into datasets by grouping data within a preset time interval. , ... ,in , , R represents the feature dimension matrix, p represents the number of samples, and q represents the dimension of the micro-meteorological features. This represents a dimension of the line's tensile strength characteristics; The raw data acquired by the sensor is reconstructed using an encoder-decoder approach. The encoding process of the raw data X acquired by the sensor from the input layer to the hidden layer includes: Where h represents a latent variable, and For the encoder weight bias, Here, X represents the activation function of the encoding layer, where X represents either data sample X1 or X2 in dataset D. The data decoding process from the hidden layer to the output layer includes: in, Data representing reconstruction, and For the weight bias of the decoder, The activation function for the decoding layer; The objective function of the autoencoder algorithm is: Where dist represents the vector distance calculation function.

4. The second-order integrated icing prediction method based on multi-source icing monitoring data according to claim 3, characterized in that, The method of establishing a prediction model for icing sequence data using features extracted from time series data includes: By using a long short-term memory network and training and debugging the model with training and test sets, a predictive model for icing sequence data is established. in, This represents the time-series characteristics of icing. This represents the mapping relationship between icing thickness and icing time-series characteristic data. This represents the output of the time series prediction model.

5. The second-order integrated icing prediction method based on multi-source icing monitoring data according to claim 1, characterized in that, The method of establishing a prediction model for icing image data based on features extracted from transmission line icing images includes: Features of icing images were extracted using the ResNet-18 model, and then the network model was trained to build a prediction model for icing image data. in, This represents the feature data of the icing image. This indicates the mapping relationship between ice thickness and ice image data. This represents the output of the image data prediction model.

6. The second-order integrated icing prediction method based on multi-source icing monitoring data according to claim 1, characterized in that, The weighting based on the similarity between micro-meteorological data and micro-meteorological data of typical micro-topography in historical statistics includes: The micro-meteorological data of the target route are clustered with typical micro-meteorological data in historical statistics to obtain multiple clusters, each cluster corresponding to a cluster center; Map the target route data to these cluster centers and calculate the Euclidean distance between them; The Euclidean distance from the target route data to a cluster center is used to characterize the similarity between the target route data and historical micro-meteorological data. Weights are then assigned to each micro-topographic meteorological scenario model based on this similarity. The calculation method for assigning weights to each micro-topographic meteorological scenario model includes: Where di represents the similarity between the target route data and the i-th micro-topography meteorological scenario model, i.e., the Euclidean distance; n represents the number of micro-topography meteorological scenario models; This represents the sum of similarity scores between each of the n micro-topographic meteorological scenario models and the target route data; The coordinates of the i-th cluster center are represented by x and y; the coordinates of the target line data are represented by x and y.

7. The second-order integrated icing prediction method based on multi-source icing monitoring data according to claim 6, characterized in that, The process involves establishing multiple multi-source data integration and prediction models under various micro-topographic meteorological scenarios, and obtaining the final icing thickness prediction result by adaptively weighted ensemble integrating information from multiple models. This includes: in, This indicates the final predicted icing thickness.

8. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing a computer program, which is loaded and executed by the processor to implement the second-order integrated icing prediction method based on multi-source icing monitoring data as described in any one of claims 1 to 7.

Images