Power transmission line icing galloping prediction method and system based on improved time sequence large model, equipment and storage medium

Abstract

The invention discloses a power transmission line icing galloping prediction method and system based on an improved time sequence large model, equipment and a storage medium, and relates to the technical field of power system safety monitoring, and the method comprises the steps: obtaining line parameters, and meteorological data and power transmission line icing galloping data in an abnormal accumulation time period; constructing a data feature project by using the line parameters, the meteorological data in the abnormal accumulation time period and the icing galloping data of the power transmission line; performing data synthesis on the meteorological data and the icing galloping data of the power transmission line in the abnormal accumulation time period through time alignment to obtain a comprehensive time sequence; and taking the comprehensive time sequence as input, and obtaining the icing galloping probability and the galloping amplitude of the power transmission line based on the output of the improved time sequence large model. According to the invention, the icing galloping risk early warning of the power transmission line can be rapidly and accurately output, and reliable guarantee is provided for safe operation of a power grid.

Description

Technical Field

[0001] The present invention relates to a method and system, equipment and storage medium for predicting ice dancing of a transmission line based on an improved time series large model, and belongs to the technical field of power system safety monitoring. Background Art

[0002] Ice-covered conductors and ground wires on high-voltage transmission lines pose a significant threat to the safe operation of power grids. In areas of low temperature, high humidity, or high altitude, ice-covered conductors trigger large, low-frequency vibrations due to wind loads. The ice layer significantly increases conductor loads and alters aerodynamic properties, leading to chain reactions such as conductor breakage, phase-to-phase flashover, and tower failure. These events pose a significant challenge to power grid stability and socioeconomic security.

[0003] Currently, ice-drift prediction relies primarily on traditional statistical regression methods, empirical formulas, and low-dimensional nonlinear models. These methods are often based on limited historical data and empirical parameters. Limited by the sparsity of characteristic dimensions, they struggle to effectively capture the complex nonlinear coupling relationships between meteorological conditions, ice accumulation, and structural responses. Furthermore, the ice-drift process is significantly influenced by the time-varying nature of external environmental factors, making traditional methods significantly inadequate in capturing dynamic changes, providing real-time responses, and ensuring robustness. Summary of the Invention

[0004] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a transmission line ice dancing prediction method and system, equipment, and storage medium based on an improved time series large model, which can quickly and accurately output transmission line ice dancing risk warnings and provide reliable protection for the safe operation of the power grid.

[0005] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0006] In a first aspect, the present invention provides a method for predicting ice dancing on a transmission line based on a time series large model, comprising:

[0007] Get real-time weather data;

[0008] The real-time meteorological data is compared with the preset threshold conditions. If the threshold conditions are not met, it is judged that there is no ice galloping risk; if the threshold conditions are met, the accumulated time of meteorological data anomalies is monitored, and the accumulated time of meteorological data anomalies is compared with the set threshold. If it is lower than the set threshold, it is judged that the ice galloping risk is lifted; if it is higher than the set threshold, the transmission line ice galloping probability and transmission line galloping amplitude are predicted based on the improved time series large model, line parameters, meteorological data within the accumulated time period of anomalies, and transmission line ice galloping data.

[0009] Furthermore, the method of predicting the probability of ice galloping of transmission lines and the amplitude of ice galloping of transmission lines based on the improved time series large model, line parameters, meteorological data within the abnormal accumulation period, and ice galloping data of transmission lines includes:

[0010] Obtain line parameters, meteorological data within the abnormal accumulation period, and ice dancing data on transmission lines;

[0011] Data feature engineering is constructed using line parameters, meteorological data within the abnormal accumulation period, and ice-covered transmission line movement data;

[0012] Based on the data feature engineering, the data features of meteorological data and transmission line ice dancing data are synthesized by time alignment, and the data features of line parameters are combined to obtain a comprehensive time series;

[0013] Taking the comprehensive time series as input, the probability of ice-covered transmission line dancing and the transmission line dancing amplitude are obtained based on the output of the improved time series large model.

[0014] Furthermore, the meteorological data includes region, time, weather, temperature, precipitation, wind direction, wind force, wind speed, air pressure, humidity, air quality, and visibility;

[0015] The line parameters include line name, voltage level, number of splits, conductor model, hanging point height, span, and conductor direction;

[0016] The transmission line ice dancing data includes line ice thickness, dancing frequency and dancing amplitude.

[0017] Furthermore, the data feature engineering is constructed by utilizing line parameters, meteorological data within the abnormal accumulation period, and ice dancing data of transmission lines, including:

[0018] Performing data preprocessing on line parameters, meteorological data within an abnormal accumulation period, and transmission line ice dancing data to obtain preprocessed meteorological data, line parameters, and transmission line ice dancing data;

[0019] Taking each data as a data feature, the Pearson correlation coefficient is used to calculate the correlation coefficient between it and the ice dancing state of the transmission line;

[0020] Data features with high correlation are screened out based on the correlation coefficient, and then standardized and combined to obtain data feature engineering.

[0021] Furthermore, the data preprocessing includes: deleting outliers and erroneous values ​​in the data, and filling missing data using mean filling and interpolation methods;

[0022] The standardization process is one of normalization, logarithmic transformation normalization, Z-Score normalization, periodic coding, directional coding, and Sigmoid transformation.

[0023] Furthermore, based on the data feature engineering, the data features of meteorological data and transmission line ice dancing data are synthesized by time alignment, and the data features of line parameters are combined to obtain a comprehensive time series, including:

[0024] The data features of meteorological data and transmission line ice dancing data are matched according to the timestamps to obtain the meteorological sequence;

[0025] The data features of the line parameters and the meteorological series are combined to generate a comprehensive time series, where the dimension of the comprehensive time series is expressed as time step × feature dimension.

[0026] Furthermore, the improved time series large model is obtained by improving the time series large model based on the Transformer, specifically comprising: replacing the output layer of the Transformer-based time series large model with two parallel task heads to obtain the improved time series large model, wherein one task head is a classification task output layer, and the other task head is a regression task output layer;

[0027] The classification task output layer includes an attention pooling layer, a fully connected layer and a Sigmoid activation function arranged in sequence;

[0028] The attention pooling layer is used to extract time step information from the feature vector output by the Transformer decoder, calculate the attention weight for each time step information, and perform weighted summation of the features of each time step to obtain a global feature vector;

[0029] The fully connected layer is used to perform linear transformation on the global feature vector;

[0030] The Sigmoid activation function is used to calculate and output a probability value on the global eigenvector after the linear transformation. The probability value range is [0, 1]. When the probability value is ≤0.5, it is determined that the transmission line is ice-covered and dancing; otherwise, it is determined that the transmission line is not ice-covered and dancing.

[0031] The regression task output layer includes a fully connected layer, which is used to map the feature vector output by the Transformer decoder through the fully connected layer to obtain a continuous value, and the continuous value represents the vibration amplitude of the ice covering the transmission line.

[0032] Furthermore, the data processing expression of the attention pooling layer is:

[0033]

[0034]

[0035] in, Indicates the The attention weight of time steps, represents the learnable weight vector, Indicates that the Transformer decoder is The output feature vector of time steps, express and The inner product of Indicates that the Transformer decoder is The output characteristic vector of each time step affects the ice dancing of the transmission line. It represents the sum of the influence of the output feature vectors of the Transformer decoder at all time steps on the ice dancing of the transmission line. represents the number of time steps, Represents the feature vector output by the attention pooling layer, that is, the global feature vector;

[0036] The calculation expression of the Sigmoid activation function is:

[0037] , in, represents the probability of ice dancing on the transmission line, represents the classifiable weight matrix, represents the global eigenvector after linear transformation, represents the bias term, Represents the Sigmoid activation function.

[0038] Furthermore, the method further includes pre-training the improved time series large model, and the pre-training method includes:

[0039] Obtain historical meteorological data, line parameters, and historical transmission line ice dancing data;

[0040] Use historical meteorological data, line parameters, and historical transmission line ice-covered movement data to build data feature engineering;

[0041] Based on data feature engineering, the data features of historical meteorological data and historical transmission line ice dancing data are synthesized through time alignment, and the data features of line parameters are combined to obtain a comprehensive time series.

[0042] The comprehensive time series is used as input to train the improved time series large model. The loss function is calculated during the training process until the loss function is less than a preset threshold, and a pre-trained improved time series large model is obtained.

[0043] Furthermore, the loss function adopts a joint loss function, wherein the classification task output layer adopts a binary cross entropy loss function, and the regression task output layer adopts a mean square error loss function;

[0044] The expression of the joint loss function is:

[0045]

[0046] in, represents the joint loss function, represents the binary cross entropy loss function, represents the mean square error loss function, 、 They represent the weights of the binary cross entropy loss function and the mean square error loss function respectively.

[0047] In a second aspect, the present invention further provides a transmission line ice dancing prediction system based on an improved time series large model, comprising:

[0048] A data acquisition module is configured to acquire real-time meteorological data;

[0049] The meteorological mutation judgment module is configured to compare the real-time meteorological data with the preset threshold conditions. If the threshold conditions are not met, it is judged that there is no risk of ice dancing; if the threshold conditions are met, the abnormal cumulative time of meteorological data is monitored, and the abnormal cumulative time of meteorological data is compared with the set threshold. If it is lower than the set threshold, it is judged that the ice dancing risk is lifted; if it is higher than the set threshold, the transmission line ice dancing probability and transmission line dancing amplitude are predicted based on the improved time series large model, line parameters, meteorological data within the abnormal cumulative time period, and transmission line ice dancing data.

[0050] Furthermore, the meteorological mutation judgment module specifically includes a data acquisition unit, a data feature engineering construction unit, a time series generation unit and an ice dance unit.

[0051] Furthermore, the data acquisition unit is configured to acquire line parameters, meteorological data within an abnormal accumulation period, and ice dancing data of the transmission line.

[0052] The data feature engineering construction unit is configured to construct data feature engineering using line parameters, meteorological data within an abnormal accumulation time period, and ice dance data of transmission lines.

[0053] Furthermore, the line parameters include line name, voltage level, number of splits, conductor model, hanging point height, span, conductor direction, etc., and the transmission line ice dancing data includes line ice thickness, dancing frequency and dancing amplitude.

[0054] Furthermore, the data feature engineering using line parameters, meteorological data within the abnormal accumulation period, and ice dancing data of transmission lines includes:

[0055] Data preprocessing is performed on the line parameters, meteorological data within the abnormal accumulation period, and transmission line ice dancing data to obtain preprocessed meteorological data, line parameters, and transmission line ice dancing data. The data preprocessing method is to delete outliers and error values ​​in the data and use mean filling and interpolation methods to fill in missing data.

[0056] Each preprocessed data item is taken as a data feature, and the Pearson correlation coefficient is used to calculate the correlation coefficient between it and the ice dancing state of the transmission line.

[0057] According to the correlation coefficient, data features with high correlation are screened out, and after standardization, they are combined to obtain data feature engineering. The standardization process can be performed by using one of normalization, logarithmic transformation normalization, Z-Score normalization, periodic coding, directional coding, and Sigmoid transformation.

[0058] Furthermore, the time series generation unit is configured to perform data synthesis on the data features of meteorological data and transmission line ice dancing data through time alignment based on the data feature engineering, and obtain a comprehensive time series by combining the data features of the line parameters. The dimension of the comprehensive time series is expressed as time step × feature dimension.

[0059] The data features of meteorological data and transmission line ice dancing data are matched according to the timestamps to obtain the meteorological sequence;

[0060] The data characteristics of line parameters and meteorological series are combined to generate a comprehensive time series.

[0061] Furthermore, the ice dancing prediction unit is configured to take the comprehensive time series as input and obtain the transmission line ice dancing probability and transmission line dancing amplitude based on the output of the improved time series large model.

[0062] The improved time series model is based on the Transformer-based time series model. The output layer of the Transformer-based time series model is replaced with two parallel task heads to obtain the improved time series model. One task head is the classification task output layer, and the other task head is the regression task output layer. Specifically:

[0063] Furthermore, the classification task output layer includes an attention pooling layer, a fully connected layer, and a Sigmoid activation function, which are set in sequence.

[0064] Among them, the attention pooling layer is used to extract time step information from the feature vector output by the Transformer decoder, calculate the attention weight for each time step information, and perform weighted summation of the features of each time step to obtain the global feature vector. Its processing expression is:

[0065]

[0066]

[0067] in, Indicates the The attention weight of time steps, represents the learnable weight vector, Indicates that the Transformer decoder is The output feature vector of time steps, express and The inner product of Indicates that the Transformer decoder is The output characteristic vector of each time step affects the ice dancing of the transmission line. It represents the sum of the influence of the output feature vectors of the Transformer decoder at all time steps on the ice dancing of the transmission line. represents the number of time steps, Represents the feature vector output by the attention pooling layer, that is, the global feature vector.

[0068] The fully connected layer is used to perform linear transformation on the global feature vector to obtain the linearly transformed global feature vector.

[0069] The Sigmoid activation function is used to calculate the global eigenvector after linear transformation and output a probability value. The probability value range is [0, 1]. When the probability value is ≤ 0.5, it is judged that the transmission line is ice-covered and dancing. Otherwise, it is judged that the transmission line is not ice-covered and dancing. Its calculation expression is:

[0070] , in, represents the probability of ice dancing on the transmission line, represents the classifiable weight matrix, represents the global eigenvector after linear transformation, represents the bias term, Represents the Sigmoid activation function.

[0071] Furthermore, the regression task output layer includes a fully connected layer, which is used to map the feature vector output by the Transformer decoder through the fully connected layer to obtain a continuous value, and the continuous value represents the vibration amplitude of the ice covering the transmission line.

[0072] In a third aspect, the present invention further provides an electronic device, comprising:

[0073] memory for storing computer programs;

[0074] A processor is used to execute the computer program to implement the steps of the method for predicting ice dancing of transmission lines based on the improved time series large model as described in any one of the first aspects.

[0075] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method for predicting ice dancing of transmission lines based on an improved time series large model as described in any one of the first aspects are implemented.

[0076] Compared with the prior art, the present invention has the following beneficial effects:

[0077] The present invention combines real-time meteorological data, line status information and the risk of ice-covered transmission lines, and obtains the probability and amplitude of ice-covered transmission lines based on the improved time series large model, breaking through the limitations of traditional single-task modeling. It can realize timely warning of the risk of ice-covered transmission lines, and capture complex time dependencies and periodic patterns through the time series reconstruction of the improved time series large model and the Transformer dual-task transformation, which greatly improves the prediction accuracy and can effectively reduce the risk of ice disasters in the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 Schematic diagram of a flow chart of a method for predicting ice dancing on a transmission line based on an improved time series large model in one embodiment of the present invention;

[0079] Figure 2 A schematic diagram of a comprehensive time series in a method for predicting ice dancing on a transmission line based on an improved time series large model in one embodiment of the present invention;

[0080] Figure 3 Schematic diagram of the architecture of the improved time series large model in the method for predicting ice dancing of transmission lines based on the improved time series large model in an embodiment of the present invention. DETAILED DESCRIPTION

[0081] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0082] Example 1:

[0083] like Figure 1 As shown, an embodiment of the present invention provides a method for predicting ice dancing of transmission lines based on an improved time series large model, which specifically includes the following steps:

[0084] Acquire real-time meteorological data. In this embodiment, the real-time meteorological data includes region, time, weather, precipitation, wind direction, wind force, wind speed, air pressure, humidity, air quality, and visibility, and is acquired through meteorological monitoring equipment.

[0085] In this embodiment, the preset threshold conditions are: temperature ≤ 0°C, humidity ≥ 80%, wind speed 5-15 m / s (force 3-7), and continuous rain and snow (such as sleet or freezing rain), with a duration of ≥ 30 minutes. Real-time meteorological data is compared with the preset threshold conditions. If the threshold conditions are not met, it is determined that there is no ice galloping risk. If the threshold conditions are met, the accumulated time of meteorological data anomalies is monitored and compared with a set threshold (set to 2 hours in this embodiment). If it is below the set threshold, it is determined that the ice galloping risk has been eliminated. If it is above the set threshold, the pre-trained improved time series large model is used to predict the probability and amplitude of ice galloping on the transmission line.

[0086] Obtain line parameters, meteorological data within the abnormal accumulation period, and transmission line ice dancing data.

[0087] Data feature engineering is constructed using line parameters, meteorological data within the abnormal accumulation period, and ice-covered transmission line movement data. Specifically:

[0088] For example, the micro-meteorological station mounted on the tower is used as the data source, with a sampling interval of 1 minute to collect wind speed (0~30m / s), wind direction, temperature (-30℃±15℃), humidity (0~100%RH) and precipitation (0~50mm / h) parameters; the conductor inclination (±30°), tension (0~50kN), vibration frequency (0~50Hz) and other parameters are obtained at a sampling frequency of 10Hz.

[0089] The data was preprocessed, including data cleaning. Specifically, outliers and erroneous data were removed. Data with amplitude values ​​significantly deviating from the normal range were deleted. Missing data were handled using mean filling and interpolation. If data was missing at a certain moment, but the data at adjacent moments was relatively stable, the average value of the adjacent data was used to fill the missing data.

[0090] The Pearson correlation coefficient is used to calculate the correlation coefficient between each feature and the conductor ice dancing state, and highly correlated features are selected. In this embodiment, highly correlated features include: region, time, weather, temperature, precipitation, wind direction, wind force, wind speed, air pressure, humidity, air quality, and visibility for meteorological data; line name, voltage level, number of splits, conductor model, hanging point height, span, and conductor orientation for line parameters; and line ice thickness, dancing frequency, and dancing amplitude for transmission line ice dancing data.

[0091] Standardize the selected highly relevant features. Feature data can be diverse in source, complex in type, at varying scales, and have strong spatiotemporal correlations. Therefore, different processing methods are required for different features to unify the data scale and enhance the model's ability to identify key factors such as periodicity and spatial direction. Standardization methods include, but are not limited to, normalization, logarithmic transformation normalization, Z-score normalization, periodic encoding, directional encoding, and sigmoid transformation.

[0092] The normalization processing methods corresponding to the various features in this embodiment are shown in Table 1 and Table 2:

[0093] Table 1: Processing methods for some highly correlated features in meteorological data

[0094]

[0095] Table 2: Processing methods for some highly correlated features in line parameters and transmission line ice dancing data

[0096]

[0097] Combining the standardized features results in data feature engineering.

[0098] Next, based on data feature engineering, the discrete wire galloping data and continuous meteorological monitoring data are synthesized through time alignment. Then, the data features of the line parameters in the data feature engineering are combined to generate a comprehensive time series. Specifically, the following are performed:

[0099] For each transmission line, the discrete dancing data and the continuous meteorological data are timestamp-matched to obtain a meteorological sequence based on the recorded area. If the sequence is within the duration of the dancing, the dancing amplitude, dancing frequency, and ice thickness are filled with the recorded data; otherwise, they are filled with 0. At the same time, the data features of the line parameters are added to the sequence. This is because the line parameters are static indicators, that is, the line parameters do not change with time. Ultimately, a comprehensive time series is generated, and each transmission line corresponds to one comprehensive time series.

[0100] The dimension of the comprehensive time series is [time step × feature dimension], such as Figure 2 As shown, the characteristic dimensions in this embodiment include 22 items: region, time, weather, temperature, precipitation, wind direction, wind force, wind speed, air pressure, humidity, air quality, visibility, line name, voltage level, number of splits, conductor model, hanging point height, span, conductor direction, line ice thickness, dancing frequency and dancing amplitude.

[0101] Next, build an improved time series model:

[0102] The time series large model in this embodiment is improved based on the time series large model of Transformer, and the output layer of the existing Transformer-based time series large model is replaced by two parallel task heads, where one task head is the classification task output layer and the other task head is the regression task output layer.

[0103] Combine Figure 3 The basic architecture of the improved large-scale time series model includes an input layer, which is used to input comprehensive time series data (time step × feature dimension). The output of the input layer is connected to a feature embedding layer and a position encoding layer. The feature embedding layer is used to project the input data into a high-dimensional space through a linear transformation to obtain a feature embedding vector. Its function is to convert physical quantities of different dimensions into vector representations of uniform dimensions while preserving the correlation between features. The position encoding layer is used to inject time series information through sine / cosine position encoding, generating a unique position vector for each time step, enabling the model to recognize time series relationships.

[0104] The position vector and the feature embedding vector are added element by element and then input into the Transformer encoder and Transformer decoder in sequence. Both the Transformer encoder and the Transformer decoder include multiple layers. Each layer of the Transformer encoder includes a multi-head attention mechanism unit and a feedforward network set in sequence. Each layer of the Transformer decoder includes a multi-head causal self-attention mechanism unit, a cross attention mechanism unit and a feedforward network set in sequence.

[0105] The classification task output layer includes an attention pooling layer, a fully connected layer, and a Sigmoid activation function, which are set in sequence. The attention pooling layer is used to extract time step information from the feature vector output by the Transformer decoder and calculate the attention weight for each time step information. The features of each time step are weighted and summed to obtain a global feature vector. The global feature vector is then linearly transformed through the fully connected layer. Finally, the Sigmoid activation function is used to operate on the linearly transformed global feature vector to output a probability value. The probability value range is [0, 1]. When the probability value is ≤0.5, it is judged that the transmission line is ice-covered and dancing; otherwise, it is judged that the transmission line is not ice-covered and dancing.

[0106] The output layer of the regression task includes a fully connected layer, which is used to map the feature vector output by the Transformer decoder into a continuous value through the fully connected layer. The continuous value represents the vibration amplitude of the ice covering the transmission line.

[0107] In this example, the improved time series large model uses Google's TimesFM (timesfm-1.0-200m) as the base model, with 200m model parameters. TimesFM's model structure is designed to handle zero-shot prediction tasks for multi-domain, multi-granularity time series. Its core architecture consists of an input layer, feature embedding layer, positional encoding layer, stacked Transformer layers, and an output layer. It utilizes a specific patching strategy and masking mechanism to achieve efficient pre-training and generalization.

[0108] The input layer consists of:

[0109] (1) Patching: The input comprehensive time series is divided into non-overlapping patches of fixed length as the input tokens of the Transformer.

[0110] (2) Residual Block Processing Unit: Each patch is converted into a vector of dimension model_dim by a residual block. A binary mask is also provided to the Transformer along with the input. The binary mask is used to indicate whether the corresponding data point should be considered (0) or ignored (1). The residual block is essentially a multi-layer perceptron (MLP) with one hidden layer and one skip connection. In order for the model to capture temporal order information, the model uses sine-cosine positional encoding.

[0111] (3) Random mask unit: used to flexibly handle arbitrary context lengths and cover all possible context lengths through a random masking strategy: during training, an offset r is randomly selected and the first r time points are masked to ensure that the model adapts to inputs from 1 to the maximum context length (such as 512).

[0112] Stacked Transformer layer Stacked Transformer includes:

[0113] (1) Decoder-only architecture: It adopts a pure decoder structure and uses only causal self-attention to ensure that each output token can only pay attention to its previous input token (including itself).

[0114] (2) Layer structure: Multi-Head Causal Self-Attention: Each Transformer layer uses a multi-head causal self-attention mechanism, allowing the model to focus on different parts of the input sequence simultaneously.

[0115] The feedforward network FFN is set after the self-attention mechanism, and each layer applies a feedforward network independently to each position in the sequence. This further processes the attention information and enables the model to learn higher-level representations.

[0116] Output Layers include:

[0117] (1) Variable-length patch prediction: The output patch length h can be larger than the input patch length p, supporting prediction of longer time windows, such as 32 input patches and 128 output patches. This design reduces the number of autoregressive steps, such as predicting 256 steps only requires 2 generation times instead of 8, improving the efficiency of long sequence prediction.

[0118] (2) Adapting to multi-granularity prediction: By adjusting h and p, the model can handle different time granularities, such as hours, days, and months, without modifying the architecture. It only needs to cover multi-granularity data in pre-training.

[0119] TimesFM is a decoder-only architecture that processes input time series data through multiple self-attention layers and a feedforward network, ultimately outputting a sequence of latent vectors. This paper modifies the TimesFM model structure for the task of ice-covered dance prediction. By replacing the TimesFM model's output layer with classification and regression output layers, the improved time series model is obtained.

[0120] Pre-train the constructed improved time series model. The pre-training methods include:

[0121] First, the raw data from meteorological monitoring equipment, line status sensors and historical ice dance records are collected, preprocessed and feature extracted, and data feature engineering is constructed.

[0122] Based on data feature engineering, the data features of historical meteorological data and historical transmission line ice dancing data are synthesized through time alignment, and the data features of line parameters are combined to obtain a comprehensive time series.

[0123] The comprehensive time series is used as input to train the improved time series large model. The loss function is calculated during the training process until the loss function is less than a preset threshold, thereby obtaining a pre-trained improved time series large model. In this embodiment, the loss function adopts a joint loss function, wherein the classification task output layer adopts a binary cross entropy loss function, and the regression task output layer adopts a mean square error loss function. The expression of the joint loss function is:

[0124]

[0125] in, represents the joint loss function, represents the binary cross entropy loss function, represents the mean square error loss function, 、 They represent the weights of the binary cross entropy loss function and the mean square error loss function respectively.

[0126] Based on the pre-trained improved time series large model, the model is fine-tuned. The fine-tuning adopts linear detection fine-tuning, low-rank adaptation fine-tuning, and weight decomposition adaptation fine-tuning strategies, and the effects are comprehensively compared.

[0127] Linear Probing fine-tuning trains only the input / output and embeddings, freezing the transformer block. It uses the common features of the pre-trained model as a fixed feature extractor, training only a simple linear layer for the specific task. This method has low parameter overhead and fast updates, making it suitable for scenarios where rapid adaptation to new tasks is desired without disturbing the model's existing knowledge.

[0128] Low-Rank Adaptation (LoRA) is a weight matrix in the original network Add a low-rank fine-tuning term on top , that is, only training the low-rank matrix without changing the original large model weights, so as to achieve the effect of efficient parameter fine-tuning. The update formula is usually:

[0129] , ;

[0130] in, Represents the weight matrix after LoRA fine-tuning, is the original weight matrix in the pre-trained model; and are two low-rank matrices (low-rank factors Usually much smaller than ), which is a trainable parameter during fine-tuning and is only updated and Perform gradient update, and Keeping the state frozen significantly reduces the number of parameters that need to be tuned while reusing pre-trained knowledge. This approach can capture task-specific variations while reducing memory and computational overhead.

[0131] Directional LoRA (DoRA) with weight decomposition adaptation can be seen as a further extension of LoRA, aiming to further distinguish between "direction" and "magnitude" information. DoRA first decomposes the weights of the pre-trained model, breaking each weight matrix into a magnitude vector and a direction matrix. This decomposition allows the model to better control the weight learning process. During fine-tuning, DoRA uses LoRA for directional updates, adjusting only the parameters of the direction portion while keeping the magnitude portion unchanged. This approach reduces the number of parameters that need to be adjusted and improves the efficiency of fine-tuning. This method simplifies the task compared to traditional fine-tuning methods, which require adjusting both the magnitude and the direction. Weight decomposition analysis helps DoRA enhance the model's learning ability and training stability. This method aims to mimic the learning capabilities of full fine-tuning while avoiding any additional inference overhead.

[0132] The update formula of DoRA is:

[0133]

[0134] in, represents the weight matrix after DoRA fine-tuning, represents a trainable amplitude variable, Represents the input feature vector of a certain layer during the forward propagation of the model, express right The increment after fine-tuning is used to control the weight strength, represents the frozen initial direction matrix, Represented by the low-rank matrix and The adapter items constituted by Represents the constraint norm.

[0135] Since DoRA is closest to full parameter fine-tuning while maintaining parameter efficiency and significantly outperforms LoRA and Linear Probing, this embodiment ultimately selects DoRA as the fine-tuning strategy, which can strike a balance between performance and efficiency.

[0136] Taking the comprehensive time series as the output, the probability of ice-covered transmission line dancing and the transmission line dancing amplitude are obtained based on the output of the improved time series large model.

[0137] Example 2:

[0138] Based on Example 1, this embodiment of the present invention provides a transmission line ice dancing prediction system based on an improved time series large model, which is implemented by deploying the improved time series large model on an edge computing terminal. The system specifically includes the following steps:

[0139] First, the pre-trained improved time series large model is subjected to structured pruning, and the improved time series large model is preprocessed using pruning technology. Among them, the pruning technology evaluates the importance of each model parameter or network module, and screens and eliminates parameters that have low contribution to the prediction results or are redundant, so as to reduce the total number of model parameters and the overall model size, and reduce the computational complexity of the model.

[0140] 16-bit quantization technology is used to reduce the numerical precision of the pruned model, converting the original 32-bit floating-point weights and activation values ​​in the model into 16-bit floating-point or fixed-point numbers, reducing memory usage and inference latency.

[0141] Convert the pruned and 16-bit quantized model to the ONNX format, and select the corresponding inference framework for deployment based on the hardware characteristics of the edge computing terminal to achieve efficient real-time inference.

[0142] The system of this embodiment specifically includes:

[0143] The data acquisition module is configured to obtain real-time meteorological data. In this embodiment, the real-time meteorological data includes region, time, weather, precipitation, wind direction, wind force, wind speed, air pressure, humidity, air quality, and visibility, which are obtained through meteorological monitoring equipment.

[0144] The meteorological mutation judgment module (edge ​​deployment module) is configured to compare real-time meteorological data with preset threshold conditions. If the threshold conditions are not met, it is judged that there is no risk of ice dancing. If the threshold conditions are met, the accumulated time of meteorological data anomalies is monitored and compared with the set threshold. If it is lower than the set threshold, it is judged that the ice dancing risk has been eliminated. If it is higher than the set threshold, the probability of ice dancing and the amplitude of transmission line ice dancing are predicted based on the improved time series large model, line parameters, meteorological data within the accumulated time period of anomalies, and ice dancing data of transmission lines.

[0145] The meteorological mutation judgment module specifically includes a data acquisition unit, a data feature engineering construction unit, a time series generation unit, and an ice dance unit.

[0146] The data acquisition unit is configured to acquire line parameters, meteorological data within the abnormal accumulation time period, and ice dancing data of the transmission line.

[0147] The data feature engineering construction unit is configured to construct data feature engineering using line parameters, meteorological data within an abnormal accumulation time period, and ice dance data of transmission lines.

[0148] Line parameters include line name, voltage level, number of splits, conductor model, hanging point height, span, conductor direction, etc. Transmission line ice dancing data includes line ice thickness, dancing frequency and dancing amplitude.

[0149] Data preprocessing is performed on the line parameters, meteorological data within the abnormal accumulation time period, and transmission line ice dancing data to obtain preprocessed meteorological data, line parameters, and transmission line ice dancing data. In this embodiment, the data preprocessing method is to delete abnormal values ​​and error values ​​in the data, and use mean filling and interpolation methods to fill in the missing data.

[0150] Each preprocessed data item is taken as a data feature, and the Pearson correlation coefficient is used to calculate the correlation coefficient between it and the ice dancing state of the transmission line.

[0151] According to the correlation coefficient, data features with high correlation are screened out, and after standardization, they are combined to obtain data feature engineering. The standardization process can be performed by using one of normalization, logarithmic transformation normalization, Z-Score normalization, periodic coding, directional coding, and Sigmoid transformation.

[0152] The time series generation unit is configured to synthesize the meteorological data and the transmission line ice dancing data within the abnormal accumulation time period through time alignment to obtain a comprehensive time series, and the dimension of the comprehensive time series is expressed as time step × feature dimension.

[0153] For each transmission line, the real-time meteorological data and the real-time transmission line ice dancing data are matched according to the timestamp to obtain the meteorological sequence;

[0154] The line parameters of the transmission line are combined with the meteorological series to generate a comprehensive time series.

[0155] The ice dancing prediction unit is configured to take the comprehensive time series as input and obtain the transmission line ice dancing probability and the transmission line dancing amplitude based on the output of the improved time series large model.

[0156] The improved time series model is based on the Transformer-based time series model. The output layer of the Transformer-based time series model is replaced with two parallel task heads to obtain the improved time series model. One task head is the classification task output layer, and the other task head is the regression task output layer. Specifically:

[0157] The output layer of the classification task includes the attention pooling layer, the fully connected layer and the Sigmoid activation function set in sequence.

[0158] Among them, the attention pooling layer is used to extract time step information from the feature vector output by the Transformer decoder, calculate the attention weight for each time step information, and perform weighted summation of the features of each time step to obtain the global feature vector. Its processing expression is:

[0159]

[0160]

[0161] in, Indicates the The attention weight of time steps, represents the learnable weight vector, Indicates that the Transformer decoder is The output feature vector of time steps, express and The inner product of Indicates that the Transformer decoder is The output characteristic vector of each time step affects the ice dancing of the transmission line. It represents the sum of the influence of the output feature vectors of the Transformer decoder at all time steps on the ice dancing of the transmission line. represents the number of time steps, Represents the feature vector output by the attention pooling layer, that is, the global feature vector.

[0162] The fully connected layer is used to perform linear transformation on the global feature vector to obtain the linearly transformed global feature vector.

[0163] The Sigmoid activation function is used to calculate the global eigenvector after linear transformation and output a probability value. The probability value range is [0, 1]. When the probability value is ≤ 0.5, it is judged that the transmission line is ice-covered and dancing. Otherwise, it is judged that the transmission line is not ice-covered and dancing. Its calculation expression is:

[0164] ,

[0165] in, represents the probability of ice dancing on the transmission line, represents the classifiable weight matrix, represents the global eigenvector after linear transformation, represents the bias term, Represents the Sigmoid activation function.

[0166] The output layer of the regression task includes a fully connected layer, which is used to map the feature vector output by the Transformer decoder into a continuous value through the fully connected layer. The continuous value represents the vibration amplitude of the ice covering the transmission line.

[0167] The improved large-scale time-series model proposed in this paper maintains excellent predictive performance while achieving efficient operation on resource-constrained equipment, fully meeting the power system's demand for real-time monitoring. Upon deployment, this lightweight model can quickly and accurately generate ice dance risk warnings, providing reliable assurance for the safe operation of the power grid.

[0168] Example 3:

[0169] This embodiment further provides an electronic device, including:

[0170] memory for storing computer programs;

[0171] A processor is used to execute the computer program to implement the steps of the method for predicting ice dancing of transmission lines based on the improved time series large model described in Example 1.

[0172] Example 4:

[0173] This embodiment also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method for predicting ice dancing of transmission lines based on the improved time series large model described in Example 1 are implemented.

[0174] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk drives, CD-ROMs, optical storage devices, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention may be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0175] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0176] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0177] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0178] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0179] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

[0180] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for predicting ice dancing on transmission lines based on an improved time series large model, characterized in that: include: Get real-time weather data; Compare the real-time meteorological data with the preset threshold conditions. If the threshold conditions are not met, it is determined that there is no ice dancing risk; If the threshold condition is met, the abnormal accumulated time of meteorological data is monitored and compared with the set threshold. If it is lower than the set threshold, it is judged that the risk of icing and dancing is lifted. If it is higher than the set threshold, the probability of icing and dancing of the transmission line and the transmission line icing and dancing data within the abnormal accumulated time period are used to predict the probability of icing and dancing of the transmission line and the amplitude of icing and dancing of the transmission line.

2. The method for predicting ice dancing of transmission lines based on an improved time series large model according to claim 1 is characterized in that: The method of predicting the probability of ice galloping of transmission lines and the amplitude of ice galloping of transmission lines based on the improved time series large model, line parameters, meteorological data within the abnormal accumulation period, and ice galloping data of transmission lines includes: Obtain line parameters, meteorological data within the abnormal accumulation period, and transmission line ice dancing data; Data feature engineering is constructed using line parameters, meteorological data within the abnormal accumulation period, and ice-covered transmission line movement data; Based on the data feature engineering, the data features of meteorological data and transmission line ice dancing data are synthesized by time alignment, and the data features of line parameters are combined to obtain a comprehensive time series; Taking the comprehensive time series as input, the probability of ice-covered transmission line dancing and the transmission line dancing amplitude are obtained based on the output of the improved time series large model.

3. The method for predicting ice dancing of transmission lines based on an improved time series large model according to claim 1 is characterized in that: The meteorological data includes region, time, weather, temperature, precipitation, wind direction, wind force, wind speed, air pressure, humidity, air quality, and visibility; The line parameters include line name, voltage level, number of splits, conductor model, hanging point height, span, and conductor direction; The transmission line ice dancing data includes line ice thickness, dancing frequency and dancing amplitude.

4. The method for predicting ice dancing of transmission lines based on an improved time series large model according to claim 2 is characterized in that: The data feature engineering is constructed by using line parameters, meteorological data within the abnormal accumulation period, and ice dancing data of transmission lines, including: Performing data preprocessing on line parameters, meteorological data within an abnormal accumulation period, and transmission line ice dancing data to obtain preprocessed meteorological data, line parameters, and transmission line ice dancing data; Each pre-processed data is taken as a data feature, and the correlation coefficient between it and the ice dancing state of the transmission line is calculated using the Pearson correlation coefficient. Data features with high correlation are screened out according to the correlation coefficient, and are standardized and combined to obtain data feature engineering.

5. The method for predicting ice dancing of transmission lines based on an improved time series large model according to claim 4 is characterized in that: The data preprocessing includes: deleting abnormal values ​​and erroneous values ​​in the data, and filling missing data with mean filling and interpolation methods; The standardization process is one of normalization, logarithmic transformation normalization, Z-Score normalization, periodic coding, directional coding, and Sigmoid transformation.

6. The method for predicting ice dancing of transmission lines based on an improved time series large model according to claim 2 is characterized in that: Based on the data feature engineering, the data features of meteorological data and transmission line ice dancing data are synthesized by time alignment, and the data features of line parameters are combined to obtain a comprehensive time series, including: The data features of meteorological data and transmission line ice dancing data are matched according to the timestamps to obtain the meteorological sequence; The data features of the line parameters and the meteorological series are combined to generate a comprehensive time series, where the dimension of the comprehensive time series is expressed as time step × feature dimension.

7. The method for predicting ice dancing of transmission lines based on an improved time series large model according to claim 1 is characterized in that: The improved time series large model is obtained by improving the time series large model based on the Transformer, specifically comprising: replacing the output layer of the Transformer-based time series large model with two parallel task heads to obtain the improved time series large model, wherein one task head is a classification task output layer, and the other task head is a regression task output layer; The classification task output layer includes an attention pooling layer, a fully connected layer and a Sigmoid activation function arranged in sequence; The attention pooling layer is used to extract time step information from the feature vector output by the Transformer decoder, calculate the attention weight for each time step information, and perform weighted summation of the features of each time step to obtain a global feature vector; The fully connected layer is used to perform linear transformation on the global feature vector; The Sigmoid activation function is used to calculate and output a probability value on the global eigenvector after the linear transformation. The probability value range is [0, 1]. When the probability value is ≤0.5, it is determined that the transmission line is ice-covered and dancing; otherwise, it is determined that the transmission line is not ice-covered and dancing. The regression task output layer includes a fully connected layer, which is used to map the feature vector output by the Transformer decoder through the fully connected layer to obtain a continuous value, and the continuous value represents the vibration amplitude of the ice covering the transmission line.

8. The method for predicting ice dancing of transmission lines based on an improved time series large model according to claim 7 is characterized in that: The data processing expression of the attention pooling layer is: ， ， in, Indicates the The attention weight of time steps, represents the learnable weight vector, Indicates that the Transformer decoder is The output feature vector of time steps, express and The inner product of Indicates that the Transformer decoder is The output characteristic vector of each time step affects the ice dancing of the transmission line. It represents the sum of the influence of the output feature vectors of the Transformer decoder at all time steps on the ice dancing of the transmission line. represents the number of time steps, represents the global eigenvector; The calculation expression of the Sigmoid activation function is: ， in, represents the probability of ice dancing on the transmission line, represents the classifiable weight matrix, represents the global eigenvector after linear transformation, represents the bias term, Represents the Sigmoid activation function.

9. The method for predicting ice dancing of transmission lines based on an improved time series large model according to claim 7, characterized in that: The improved time series large model is pre-trained, and the pre-training method includes: Obtain historical meteorological data, line parameters, and historical transmission line ice dancing data; Use historical meteorological data, line parameters, and historical transmission line ice-covered movement data to build data feature engineering; Based on data feature engineering, the data features of historical meteorological data and historical transmission line ice dancing data are synthesized through time alignment, and the data features of line parameters are combined to obtain a comprehensive time series. The comprehensive time series is used as input to train the improved time series large model. The loss function is calculated during the training process until the loss function is less than a preset threshold, and a pre-trained improved time series large model is obtained.

10. The method for predicting ice dancing of transmission lines based on an improved time series large model according to claim 9, characterized in that: The loss function adopts a joint loss function, wherein the classification task output layer adopts a binary cross entropy loss function, and the regression task output layer adopts a mean square error loss function; The expression of the joint loss function is: ， in, represents the joint loss function, represents the binary cross entropy loss function, represents the mean square error loss function, 、 They represent the weights of the binary cross entropy loss function and the mean square error loss function respectively.

11. A transmission line ice dancing prediction system based on an improved time series large model, characterized by: include: A data acquisition module is configured to acquire real-time meteorological data; A meteorological mutation judgment module is configured to compare real-time meteorological data with a preset threshold condition, and if the threshold condition is not met, it is judged that there is no ice dancing risk; If the threshold condition is met, the abnormal accumulated time of meteorological data is monitored and compared with the set threshold. If it is lower than the set threshold, it is judged that the risk of icing and dancing is lifted. If it is higher than the set threshold, the probability of icing and dancing of the transmission line and the transmission line icing and dancing data within the abnormal accumulated time period are used to predict the probability of icing and dancing of the transmission line and the amplitude of icing and dancing of the transmission line.

12. An electronic device, characterized in that: include: Memory for storing computer programs; A processor is used to execute the computer program to implement the steps of the method for predicting ice dancing of transmission lines based on an improved time series large model as described in any one of claims 1 to 10.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting ice dancing of a transmission line based on an improved time series large model according to any one of claims 1 to 10 are implemented.

Images