Icing thickness prediction method, system and equipment based on grey correlation and medium

Abstract

The invention relates to the technical field of meteorological prediction and artificial intelligence, in particular to an icing thickness prediction method, system and device based on gray correlation and a medium, comprising: collecting icing multi-dimensional data, and dynamically screening key meteorological features through gray correlation; constructing a data set subjected to cleaning and normalization processing and dividing the data set; a bidirectional gating circulation unit and a multi-head self-attention mechanism are fused to design a hybrid model; performing parameter optimization and test verification based on the training set; redundant interference is eliminated through dynamic feature screening, long-term and short-term icing evolution laws are analyzed bidirectionally, extreme meteorological event feature focusing is enhanced through an attention mechanism, cross-climate-region migration deployment is achieved in combination with a parameter self-adaptive compensation module, prediction precision and ice and snow sudden change working condition early warning capacity are remarkably improved, and intelligent decision support is provided for power grid disaster prevention.

Description

Technical Field

[0001] The present invention relates to the technical fields of meteorological prediction and artificial intelligence, and particularly relates to a method, system, device and medium for predicting icing thickness based on grey correlation. Background Art

[0002] As an important natural disaster endangering the safety of the power grid, icing on transmission lines exhibits significant spatio-temporal distribution characteristics under specific climatic conditions. This phenomenon is extremely likely to occur in a low-temperature and high-humidity environment. The ice accumulation process involves complex interaction mechanisms of thermodynamics and fluid mechanics, and its destructive effects penetrate multiple physical domains such as mechanics, electricity, and aerodynamics. International case analysis shows that severe icing accidents not only cause damage to power grid facilities, but may also trigger regional power supply crises, highlighting the necessity of precise prediction technology.

[0003] The development of the research field has undergone three stages of evolution: In the early stage, the physical model based on the thermodynamic equilibrium equation, although having clear mechanisms, is extremely sensitive to micro-meteorological dynamic parameters, and its calculation efficiency is difficult to meet the requirements of real-time monitoring; traditional statistical methods are limited by the non-linear coupling relationship of meteorological parameters, with insufficient prediction accuracy and generalization ability; although machine learning technology has opened up a new path for non-linear modeling, it faces bottlenecks such as complex feature engineering and insufficient utilization of time series information.

[0004] The introduction of deep learning technology marks an innovation in methodology. Time series models represented by LSTM (Long Short-Term Memory Network) effectively capture the time dependence of the icing process through gating mechanisms, and hybrid architecture models further optimize the prediction performance through spatial-temporal feature fusion. However, the current technical system still has three limitations - low efficiency in fusing multi-source heterogeneous data, lag in response to extreme working conditions, and insufficient cross-regional adaptability of the model. These pain points stem from the scientific challenges inherent in icing prediction: the non-linear characteristics of the multi-parameter action mechanism, the cumulative effect of historical meteorological conditions, the second-order impact of meteorological mutations, and geographical and climatic differences.

[0005] Aiming at the defects of existing deep models, this study proposes an innovative architecture that integrates grey correlation analysis and bidirectional gating mechanisms. By dynamically screening key meteorological factors through grey correlation, a concise and efficient input feature set is constructed; bidirectional gating units are used to capture the forward and backward correlation characteristics of time series data, avoiding the problem of long-range dependence loss in traditional recurrent neural networks; an attention mechanism is introduced to enhance the model's real-time perception ability of the mutation characteristics of the icing rate. This method breakthroughly realizes the collaborative characterization of dynamic coupling modeling of meteorological parameters and the law of time evolution. Compared with mainstream deep learning solutions, the new model shows more robust prediction performance in multi-climate zone tests, especially having significant advantages under key working conditions such as rain-snow conversion and sudden temperature changes, providing key technical support for building an active defense power grid disaster prevention system. Summary of the Invention

[0006] In view of the problems existing in the above-mentioned prior art, the present invention is proposed.

[0007] Therefore, the technical problem to be solved by the present invention is: screening out key features from meteorological data through the grey relational analysis method, and constructing a prediction model by combining differential processing, bidirectional gated recurrent unit (BiGRU) and multi-head self-attention mechanism. First, use the grey relational degree to quantify the correlation between features and the target variable, and optimize the input feature dimension; secondly, eliminate the non-stationarity of the data through first-order difference to improve the model training efficiency; finally, adopt bidirectional GRU (gated recurrent unit) to capture the bidirectional dependence relationship of time series data, and dynamically allocate weights for different time steps through the multi-head self-attention mechanism; finally, output the prediction result through the fully connected layer.

[0008] To solve the above technical problems, the present invention provides the following technical solution. A grey relational icing thickness prediction method includes: using meteorological stations and sensors to collect icing-related data in real time in different regions and periods; using the grey relational analysis method to screen out important features with high correlation with the icing thickness from the features by using a threshold; preprocessing the collected data, constructing an icing thickness prediction data set and dividing it into a training set and a test set; combining the bidirectional gated recurrent unit and the multi-head self-attention mechanism to construct a hybrid prediction model; using the training set to train the combined model, and using the test set data for prediction to complete model testing and obtain the prediction result.

[0009] As a preferred scheme of the grey relational icing thickness prediction method described in the present invention, wherein: the collection of icing-related data includes using meteorological stations and sensor networks to conduct real-time online detection on the icing areas of transmission lines, including temperature, humidity, wind speed, ice grade, precipitation and icing data; Among them, the icing data includes the recorded ice thickness of the transmission line in the past period of time.

[0010] As a preferred scheme of the grey relational icing thickness prediction method described in the present invention, wherein: the grey relational analysis method includes calculating the comprehensive correlation degree between the input meteorological features and the icing thickness, establishing a comprehensive correlation model according to the relationship between the icing thickness and the meteorological parameters, and obtaining the grey comprehensive correlation degree.

[0011] As a preferred scheme of the grey relational icing thickness prediction method described in the present invention, wherein: screening out important features with high correlation with the icing thickness includes quantifying the correlation between features and the icing thickness from multi-source meteorological parameters by using the grey relational analysis method, arranging the obtained grey comprehensive correlation degrees from large to small, and screening out features with grey comprehensive correlation degrees higher than the set threshold.

[0012] As a preferred solution of a method for predicting icing thickness based on grey correlation according to the present invention, wherein: constructing an icing thickness prediction data set and dividing it into a training set and a test set includes using historical observed icing data and screening meteorological data with high comprehensive correlation degree as input time series, using the predicted value of future icing thickness as the target time series, and constructing an icing thickness time series data set; performing preprocessing, including extracting sample point data, data cleaning and data normalization; and dividing the data set into a training set and a test set according to a ratio.

[0013] As a preferred solution of a method for predicting icing thickness based on grey correlation according to the present invention, wherein: constructing the hybrid prediction model includes combining a bidirectional gated recurrent unit and a multi-head self-attention mechanism, using the gated recurrent unit to capture the bidirectional dependence relationship of time series data, and assigning weights to time steps through the multi-head self-attention mechanism; after the bidirectional gated recurrent unit and the multi-head self-attention mechanism are combined, a feature representation is obtained, and after passing through a fully connected layer, the result of the hybrid prediction model is obtained.

[0014] As a preferred solution of a method for predicting icing thickness based on grey correlation according to the present invention, wherein: training the combined model trains the hybrid prediction model through the training set data, inputs the data into the model for prediction through a sliding window mechanism, and adjusts the parameters through an optimizer, sets the learning rate and the number of iterations, and completes training and testing.

[0015] Another object of the present invention is to provide a system for predicting icing thickness based on grey correlation, which can obtain icing-related meteorological and conductor state data in real time through a data acquisition module, and use an association screening module to quantify the grey comprehensive correlation degree between features and icing thickness and screen key parameters; with the help of a data adjustment module, clean, normalize and time-series multi-source heterogeneous data to construct a prediction data set; through a hybrid modeling module, fuse a bidirectional gated recurrent unit and a multi-head self-attention mechanism to capture the bidirectional dependence relationship of time series data and the ability of dynamic weight allocation respectively; finally, the intelligent optimization module drives the model training and testing, combines differential processing to eliminate non-stationary interference, realizes accurate short-term prediction of icing thickness, and provides real-time and reliable decision support for power grid icing disaster prevention and control.

[0016] To solve the above technical problems, the present invention provides the following technical solutions: a system for predicting icing thickness based on grey correlation, including: a data acquisition module, an association screening module, a data adjustment module, a hybrid modeling module and an intelligent optimization module; The data acquisition module uses meteorological stations and sensors to collect icing-related data in real time in different regions and periods; The association screening module uses the grey correlation analysis method to screen out important features with high correlation degree with icing thickness from features by using a threshold; The data adjustment module preprocesses the collected data, constructs an ice thickness prediction data set, and divides it into a training set and a test set. The hybrid modeling module combines a bidirectional gated recurrent unit and a multi-head self-attention mechanism to construct a hybrid prediction model. The intelligent optimization module uses the training set to train the combined model, and uses the test set data for prediction to complete model testing and obtain the prediction result.

[0017] A computer device includes a memory and a processor. The memory stores a computer program. It is characterized in that when the processor executes the computer program, the steps of a method for predicting ice thickness based on grey correlation as described above are implemented.

[0018] A computer-readable storage medium stores a computer program. It is characterized in that when the computer program is executed by a processor, the steps of a method for predicting ice thickness based on grey correlation as described above are implemented.

[0019] Advantages of the present invention: By innovatively integrating the grey correlation dynamic screening mechanism and the bidirectional gated time series modeling technology, the present invention significantly improves the prediction efficiency and adaptability of the ice thickness of transmission lines; the grey correlation algorithm constructs a multi-dimensional meteorological parameter dynamic evaluation system, intelligently identifies the key driving factors for ice layer growth, breaks through the static limitations of traditional feature selection, and effectively reduces the interference of redundant data on the model performance; the bidirectional gated network synchronously analyzes the forward and backward evolution logics of the icing process through a double-layer recursive architecture, accurately captures the long-term and short-term dependence laws of ice thickness with meteorological evolution, and overcomes the memory decay defect of conventional recurrent networks in complex time series modeling; the introduction of the self-attention mechanism forms a multi-dimensional feature focusing channel, strengthens the model's ability to capture abnormal waveforms under sudden change conditions such as rain-snow conversion and temperature abrupt changes, and significantly improves the warning sensitivity of extreme meteorological events; by constructing a parameter adaptive compensation module, the model can autonomously learn the differences in the ice formation mechanisms of different typical climate regions, eliminate the rigid constraints of regional characteristics on the prediction model, and achieve reliable migration in complex terrain environments across latitudes; through triple iterative improvements of feature optimization, time series modeling, and generalization optimization, the ice thickness prediction is advanced from traditional single physical index analysis to the intelligent decision-making stage of multi-source fusion, providing all-weather and multi-scenario dynamic ice disaster risk assessment support for the power grid, and effectively enhancing the forward-looking defense ability of the power system against natural disasters. Description of the Drawings

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for description in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0021] Figure 1 Flow chart of a method for predicting icing thickness based on grey correlation provided by an embodiment of the present invention.

[0022] Figure 2 Grey correlation degree flow chart of a method for predicting icing thickness based on grey correlation provided by an embodiment of the present invention.

[0023] Figure 3 Grey correlation degree diagram of icing of a method for predicting icing thickness based on grey correlation provided by an embodiment of the present invention.

[0024] Figure 4 Model modeling flow chart of a method for predicting icing thickness based on grey correlation provided by an embodiment of the present invention.

[0025] Figure 5 LSTM model prediction result diagram of a method for predicting icing thickness based on grey correlation provided by an embodiment of the present invention.

[0026] Figure 6 CNN model prediction result diagram of a method for predicting icing thickness based on grey correlation provided by an embodiment of the present invention.

[0027] Figure 7 BiGRU-MSA model prediction result diagram of a method for predicting icing thickness based on grey correlation provided by an embodiment of the present invention. Detailed implementation manners

[0028] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following provides a detailed description of the specific implementation manners of the present invention with reference to the accompanying drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0029] Embodiment 1, referring to Figures 1-4 , which is the first embodiment of the present invention. This embodiment provides a method for predicting icing thickness based on grey correlation, including: S1: Use meteorological stations and sensors to collect icing-related data in real time in different regions and time periods.

[0030] It should be noted that, as Figure 1As shown in step S1, the collection of icing-related data includes real-time online detection of the icing area of the transmission line using a weather station and a sensor network, including temperature, humidity, wind speed, ice grade, precipitation, and icing data; among them, the icing data includes the recorded ice thickness of the transmission line over a period of time in the past.

[0031] Furthermore, the collection equipment uses a high-precision weather station and a sensor network to record in real time the meteorological data related to the formation and evolution of icing, including temperature sensors, humidity sensors, wind speed and direction sensors, precipitation sensors, and barometric pressure sensors; the meteorological data is recorded at hourly intervals to ensure that the real-time changes in meteorological conditions can be captured; Specifically, the temperature sensor is used to record the ambient temperature with an accuracy of ±0.1°C; the humidity sensor is used to measure the relative humidity of the air with an accuracy of ±2%; the wind speed and direction sensors record the wind speed and direction, with a wind speed accuracy of ±0.2 m / s and a wind direction accuracy of ±3°; the precipitation sensor is used to measure the precipitation with an accuracy of ±0.1 mm; the barometric pressure sensor records the atmospheric pressure with an accuracy of ±0.3 hPa.

[0032] S2: Using the grey relational analysis method, use a threshold to screen out the important features with a high degree of correlation with the ice thickness from the features.

[0033] It should be noted that, as Figure 1 shown in step S2, screening the features related to the ice thickness includes using the grey relational analysis method to quantify the correlation between the features and the ice thickness from multi-source meteorological parameters, arranging the obtained grey comprehensive correlation degrees from large to small, and screening the features with a grey comprehensive correlation degree higher than the set threshold.

[0034] Furthermore, record the collected meteorological data, with temperature denoted as , humidity , wind speed , ice grade , precipitation , and ice thickness ; use the grey relational analysis method to calculate the comprehensive correlation degree between the input meteorological features and the ice thickness. According to the relationship between the ice thickness and the meteorological parameters (temperature , humidity , wind speed , ice grade , precipitation ), establish a comprehensive correlation model, expressed as: , Among them, represents the ice thickness and the th meteorological parameter The comprehensive correlation degree between , respectively represents the grey absolute correlation degree of ice thickness and meteorological parameters , and the grey relative correlation degree, which is used to balance the roles of the absolute correlation degree and the relative correlation degree; Specifically, on the one hand, the grey comprehensive correlation degree reflects the degree of geometric shape similarity between the research sequence and the research , and on the other hand, it also characterizes the degree of closeness of the change rate of the research sequence to the research sequence relative to the initial point; For example, if the prediction target sequence is , and the relevant influencing factor sequence is ; if the lengths of the two sequences are the same, the grey absolute correlation degree of the two sequences is expressed as: , where: , , , = , , where, represents the element of the target sequence, represents the element of the relevant influencing factor sequence, represents the cumulative absolute value of the prediction target sequence , represents the cumulative absolute value of the relevant influencing factor sequence , represents the absolute value of the difference between the cumulative absolute values of the prediction target sequence and the relevant influencing factor sequence , represents the original value of the -th element in the prediction target sequence , represents the original value of the -th element in the relevant influencing factor sequence , represents the summation index, indicating the traversal from the second element to the penultimate element of the sequence; represents the length of the sequence, indicating the total number of elements in the sequence. When the target sequence is , and the relevant influencing factor sequence is When the lengths of the two sequences are the same and the initial values are not zero, and 's initial value terms can be counted as and ; Among them: , , is called and 's grey absolute correlation degree is and 's grey relative correlation degree, denoted as ; , Among them: , , , , = , Among them, represents the ratio of the th element in the predicted target sequence to the first element, represents the ratio of the th element in the related influencing factor sequence to the first element, represents the cumulative absolute value of the initial value term of the predicted target sequence, represents the cumulative absolute value of the initial value term of the related influencing factor sequence, represents the absolute value of the difference between the cumulative absolute value of the initial value term of the predicted target sequence and the cumulative absolute value of the initial value term of the related influencing factor sequence, represents the original data sequence of the predicted target sequence, that is, itself, represents the original data sequence of the related influencing factor sequence, that is, itself; In addition, the grey relative correlation degree is a characterization of the correlation of the change rates of the two sequences relative to the initial point, and is only related to the change rates of the sequences relative to the initial point, and has nothing to do with the magnitudes of the values of each sequence.

[0035] Further, calculate the specific comprehensive correlation degree values of meteorological characteristics and icing thickness according to the formula, arrange the values from large to small, and the larger the value of the important feature; select the meteorological characteristics with a gray correlation degree greater than 0.7. The gray correlation degree flow chart is as Figure 2 shown, and the gray correlation degree of icing is as Figure 3 shown, Figure 2 Among them, the thirteen characteristics in

[0036] are, in turn, average wind speed, slope, daily precipitation, altitude, minimum temperature, average temperature, ice class, maximum temperature, slope, longitude, latitude, average humidity, and slope aspect.

[0037] It should be noted that, as shown in step S3 of Figure 1 , constructing the icing thickness prediction data set and dividing the data set includes using historical observed icing data and screened meteorological data with high comprehensive correlation as the input time series, and using the predicted value of future icing thickness as the target time series to construct the icing thickness time series data set; performing preprocessing, including extracting sample point data, data cleaning, and data normalization; and dividing the data set into a training set and a test set according to a ratio.

[0038] Further, use historical observed icing data and screened meteorological data with large comprehensive correlation, such as wind speed, daily precipitation, altitude, and micro-topography, as the input time series, and use the predicted value of future icing thickness as the target time series to construct the icing thickness time series data set; and perform preprocessing, including extracting sample point data, data cleaning, and data normalization; Specifically, extract the sample points of the data set, that is, the original data generated by climate change during the icing process; collect it in real time, and use the real-time data once a day as the original sample points of the data set, as shown in Table 1; Table 1 Data set samples , Among them, for data cleaning, use the average value method to fill in the missing or abnormal sample point data in the icing data set sample points through the average value of the adjacent time period sample points at the previous moments. The average value calculation formula is: , where represents the calculated average value, represents the moment when missing data or abnormal data appears, represents the selected number of samples, represents the sample point data at the th moment before the moment, The sample point data at the th moment before a certain moment, denotes the sample point data at the 1st moment before a certain moment; Data normalization is to transform the original data and map it to a range that is defaulted to between. The calculation formula for data normalization is expressed as: , where denotes the sample point data, denotes the minimum value of the sample data, denotes the maximum value of the sample data, denotes the data after normalization processing.

[0039] Furthermore, the data set is divided into a training set and a test set in a ratio of 8:2.

[0040] S4: Combine the bidirectional gated recurrent unit and the multi-head self-attention mechanism to construct a hybrid prediction model.

[0041] It should be noted that, as shown in step S4 of Figure 1 , constructing the hybrid prediction model includes combining the bidirectional gated recurrent unit and the multi-head self-attention mechanism, using the gated recurrent unit to capture the bidirectional dependencies of the time series data, and assigning weights to the time steps through the multi-head self-attention mechanism; after combining the bidirectional gated recurrent unit and the multi-head self-attention mechanism, a feature representation is obtained, which is passed through a fully connected layer to obtain the result of the hybrid prediction model.

[0042] Furthermore, in order to improve the training efficiency of the model, the first-order difference method is used to eliminate the non-stationarity of the data; the first-order difference operation makes the original non-stationary sequence become stationary, which is helpful for the subsequent learning process of the model; training the model on the differenced data can improve the prediction accuracy and stability; Use a bidirectional GRU (gated recurrent unit) to capture the bidirectional dependencies of the time series data. The bidirectional GRU structure refers to a neural network structure composed of two GRU units at the same time: one processes the input sequence in the forward order of time (referred to as the forward GRU), and one processes the input sequence in the reverse order of time (referred to as the backward GRU); for each time step t, the forward GRU generates a hidden state , and the backward GRU generates a hidden state , and the final output state is concatenated by these two states, that is: , where the symbol "[ ; ]" means concatenating two vectors in the dimension.

[0043] The structures and calculation methods of the forward GRU and the backward GRU are the same as those of the unidirectional GRU, except that the processing order is reversed. The calculation formulas for the update gate, reset gate, and candidate state of each GRU are expressed as: , , , , Among them, the update gate and the reset gate control the update and forgetting of information, and are the weight matrices of the update gate, and are the weight matrices of the reset gate, is the Sigmoid activation function. The candidate hidden state is calculated by combining the current input and the previous hidden state after reset, represents element-wise multiplication, and are the weight matrices used to generate the candidate hidden state, represents the hyperbolic tangent activation function; the final hidden state of the current time step is generated, is the hidden state of the previous time step, which contains information from previous time steps.

[0044] Correspondingly, the hidden state sequence is output: The forward GRU processes the input sequence in the forward order of time and correspondingly outputs the hidden state sequence: , The forward GRU processes the input sequence in the reverse order of time and correspondingly outputs the hidden state sequence: .

[0045] Furthermore, the self-attention mechanism assigns a weight to each time step by calculating the similarity between different time steps, which is expressed as: , Among them, , , are the weight matrices of the query, key, and value, , , They are the matrix representations of queries, keys, and values respectively, used to calculate attention weights. The similarity between query vectors and key vectors is measured by calculating their dot products, and the similarities are normalized to weights through the softmax function. Then, these weights are used to perform weighted summation on value vectors; , Among them, is a mechanism, a technique for enhancing the model's focusing ability on specific parts of the input data, is the similarity matrix between queries and keys, is the dimension of keys, The operation ensures the normalization of weights, and the output at each position can be updated by weighting according to the information at other positions; After passing through the bidirectional GRU and the multi-head self-attention mechanism, a feature representation will be obtained, which will be passed to the fully connected layer to finally output the prediction result , expressed as: , Among them, is the feature representation after being processed by the bidirectional GRU and the multi-head self-attention mechanism, containing information of the sequence data at each time step; is the weight matrix of the fully connected layer, used to map to the final prediction space; is the bias term of the fully connected layer, used to adjust the baseline of the prediction output; The model prediction process is as Figure 5 shown.

[0046] S5: Use the training set to train the combined model, and use the test set data for prediction to complete model testing and obtain the prediction result.

[0047] It should be noted that, as shown in step S5 of Figure 1 , training the combined model is to train the hybrid prediction model through the training set data, input the data into the model for prediction through the sliding window mechanism, and adjust the parameters through the optimizer, set the learning rate and the number of iterations to complete training and testing.

[0048] Furthermore, select the operation data of multiple monitoring devices during multiple icing periods, and train a short-term prediction model for ice thickness based on grey correlation and BiGRU-MSA (a hybrid model of bidirectional gated recurrent unit and multi-head self-attention mechanism) through the training set data; the model inputs gas data, and the goal is to predict the gas content; through the sliding window mechanism, use historical data each time to predict the gas concentration change of the next day, and gradually obtain the future gas prediction data; the entire model is implemented using the PyTorch (deep learning) framework, set the number of neurons in the hidden layer, the optimizer uses the Adam (Adaptive Moment Estimation) algorithm, set the learning rate, and perform multiple iterations during the training process.

[0049] Example 2, refer to Figures 5-7 , the second embodiment of the present invention provides a method for predicting ice thickness based on grey correlation. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0050] Select the operation data of 625 days of icing period collected by multiple monitoring devices. As shown in Table 2, the input features of the model are the gas data of the previous 10 days, and the goal is to predict the gas content on the 11th day. Through the sliding window mechanism, use the data of the previous 10 days each time to predict the gas concentration change of the next day, and gradually obtain the future gas prediction data. The entire model is implemented using the PyTorch framework. The number of neurons in the hidden layer of this model is set to 64, the optimizer uses the Adam algorithm, the learning rate is 0.02, and the training process is carried out for 150 iterations in total.

[0051] Table 2 Operation data of 625 days of icing period collected by monitoring devices , The prediction effect obtained by using the classical model LSTM is as Figure 5 shown. The prediction accuracy is 33.00%. In the figure, TrueValue is the true value, Predicted Value is the predicted value, Value is the ice thickness value, Best Epoch is the round with the best prediction effect, True vs Predicted Values is the true value relative to the predicted value, and Time is the time.

[0052] The prediction effect obtained by using the classical model BiGRU Figure 6 is shown. The prediction accuracy is 36.23%. In the figure, TrueValue is the true value, Predicted Value is the predicted value, Value is the ice thickness value, Best Epoch is the round with the best prediction effect, True vs Predicted Values is the true value relative to the predicted value, and Time is the time.

[0053] After completing the model training and comparing the actual data with the predicted results, the comparison graph is as follows Figure 7 shown. In the figure, TrueValue is the true value, Predicted Value is the predicted value, Value is the ice coating thickness value, Best Epoch is the epoch with the best prediction effect, True vs Predicted Values is the true value relative to the predicted value, and Time is the time.

[0054] It can be seen from the prediction results that the coincidence degree between the predicted value and the actual value is relatively high. The prediction accuracy is 84.16%, MAE is 5.1769, MSE is 49.1601, RMSE is 7.0114, and MAPE is 11.99%. This indicates that the short-term prediction model for ice coating thickness based on grey correlation and BiGRU-MSA has much higher prediction accuracy than the classical model and can be used in ice coating thickness monitoring devices and systems to improve the measurement accuracy and guide the decision-making of ice coating disposal.

[0055] Example 3 is the third example of the present invention. What is different from the previous two examples is that If the described function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs and other various media that can store program codes.

[0056] The logic and / or steps described in other ways herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch instructions from the instruction execution system, apparatus, or device and execute the instructions), or in combination with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device.

[0057] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0058] Embodiment 4 is the fourth embodiment of the present invention. This embodiment provides a grey correlation icing thickness prediction system, including: a data acquisition module, a correlation screening module, a data adjustment module, a hybrid modeling module, and an intelligent optimization module; The data acquisition module uses meteorological stations and sensors to collect icing-related data in real time in different regions and periods; The correlation screening module uses the grey correlation analysis method to screen out important features with a high correlation degree with the icing thickness from the features using a threshold; The data adjustment module preprocesses the collected data, constructs an icing thickness prediction data set and divides it into a training set and a test set; The hybrid modeling module combines a bidirectional gated recurrent unit and a multi-head self-attention mechanism to construct a hybrid prediction model; The intelligent optimization module uses the training set to train the combined model, and uses the test set data for prediction to complete model testing and obtain a prediction result.

[0059] This embodiment also provides a computing device applicable to a situation of a grey correlation icing thickness prediction method, including: A memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement a grey correlation icing thickness prediction method as proposed in the above embodiment.

[0060] This embodiment also provides a storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements a grey correlation icing thickness prediction method as proposed in the above embodiment.

[0061] The storage medium proposed in this embodiment and a grey correlation icing thickness prediction method proposed in the above embodiment belong to the same inventive concept. Technical details not described in detail in this embodiment can be referred to in the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0062] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting ice coating thickness based on grey correlation, characterized in that: Including: Collecting icing-related data in real time using weather stations and sensors in different regions and time periods; Using the grey relational analysis method and a threshold to screen out important features with a high correlation degree with the icing thickness from the features; Preprocessing the collected data, constructing an icing thickness prediction data set and dividing it into a training set and a test set; Combining a bidirectional gated recurrent unit and a multi-head self-attention mechanism to construct a hybrid prediction model; Using the training set to train the combined model, and using the test set data for prediction, completing model testing, and obtaining the prediction result.

2. The method for predicting icing thickness based on grey correlation according to claim 1, wherein: The collection of icing-related data includes real-time online detection of the icing area of the transmission line using a weather station and a sensor network, including temperature, humidity, wind speed, ice grade, precipitation, and icing data; Among them, the icing data includes the recorded ice accumulation thickness of the transmission line in the past period of time.

3. The method for predicting icing thickness based on grey correlation according to claim 2, wherein: The grey relational analysis method includes calculating the comprehensive correlation degree between the input meteorological features and the icing thickness, establishing a comprehensive correlation model based on the relationship between the icing thickness and the meteorological parameters, and obtaining the grey comprehensive correlation degree.

4. The ice accretion thickness prediction method based on grey correlation according to claim 3, characterized in that: The screening of important features with a high correlation degree with the icing thickness includes quantifying the correlation between the features and the icing thickness from multi-source meteorological parameters using the grey relational analysis method, arranging the obtained grey comprehensive correlation degrees from large to small, and screening out the features with a grey comprehensive correlation degree higher than the set threshold.

5. The ice accretion thickness prediction method based on grey correlation as claimed in claim 4, wherein: The construction of the icing thickness prediction data set and its division into a training set and a test set include using the historical observed ice accumulation data and the meteorological data with a high screening comprehensive correlation degree as the input time series, and using the predicted value of the future icing thickness as the target time series to construct an icing thickness time series data set; Performing preprocessing, including extracting sample point data, data cleaning, and data normalization; and dividing the data set into a training set and a test set according to a ratio.

6. The method for predicting icing thickness based on grey correlation according to claim 5, wherein: The construction of the hybrid prediction model includes combining a bidirectional gated recurrent unit and a multi-head self-attention mechanism, using the gated recurrent unit to capture the bidirectional dependence relationship of the time series data, and assigning weights to the time steps through the multi-head self-attention mechanism; after the combination of the bidirectional gated recurrent unit and the multi-head self-attention mechanism, a feature representation is obtained, which is passed through a fully connected layer to obtain the result of the hybrid prediction model.

7. The method for predicting icing thickness based on grey correlation according to claim 6, wherein: The training of the combined model trains the hybrid prediction model through the training set data, inputs the data into the model for prediction through the sliding window mechanism, and adjusts the parameters through the optimizer, sets the learning rate and the number of iterations, and completes the training and testing.

8. A system for predicting icing thickness based on grey correlation, which applies a method for predicting icing thickness based on grey correlation as described in any one of claims 1 to 7, characterized in that: Including: A data collection module, a correlation screening module, a data adjustment module, a hybrid modeling module, and an intelligent optimization module; The data collection module collects icing-related data in real time using weather stations and sensors in different regions and time periods; The correlation screening module uses the grey relational analysis method and a threshold to screen out important features with a high correlation degree with the icing thickness from the features; The data adjustment module preprocesses the collected data, constructs an icing thickness prediction data set and divides it into a training set and a test set; The hybrid modeling module combines a bidirectional gated recurrent unit and a multi-head self-attention mechanism to construct a hybrid prediction model; The intelligent optimization module uses the training set to train the combined model, and uses the test set data for prediction to complete the model test and obtain the prediction result.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of a method for predicting icing thickness based on grey correlation according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of a method for predicting icing thickness based on grey correlation according to any one of claims 1 to 7.

Images