A power line icing simulation and prediction method based on data assimilation and artificial intelligence

Abstract

The application provides a power line icing simulation and prediction method based on data assimilation and artificial intelligence, comprising the following steps: step 1, meteorological elements affecting power line icing generation are predicted; step 2, the initial field of the meteorological mode prediction is adjusted, and the result of the numerical mode prediction in step 1 is optimized; step 3, the result of the numerical mode prediction after the optimization in step 2 is subjected to bias correction processing; step 4, whether icing is generated is determined, and the geographical position where power line icing is likely to occur is marked to form icing spatial distribution information field of the concerned area; and step 5, the warning result of the regional icing thickness is output. The application uses an artificial intelligence integrated constraint model to correct the meteorological prediction data field, obtains corrected meteorological information, calculates the icing spatial distribution field by using a mathematical method, and outputs the icing thickness prediction result, thereby assisting in solving the power line icing warning problem.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power line icing prediction, and particularly relates to a power line icing simulation prediction method based on data assimilation and artificial intelligence. BACKGROUND

[0002] Power line icing can increase the load of power line conductors and towers, expand the windward area, cause unstable oscillation of the conductors, and cause events such as pole collapse, wire breakage, twisting, and flashover, thereby causing power accidents.

[0003] Power lines may fail due to abnormal icing events. Power line icing has an important influence on the safe operation of the power transmission network, and power line icing thickness prediction helps to assess line icing risks. Therefore, to effectively reduce icing disasters during the operation of power lines, it is important to conduct high-resolution early warning in areas prone to icing. The accuracy of regional icing prediction is low at home and abroad at present, which cannot meet the needs of power grid and other related departments for ice prevention and disaster reduction. The present application hopes to introduce meteorological elements to improve the accuracy of future icing trend prediction of power lines, and to provide theoretical and application support for disaster prevention and reduction of power line, mountain wind power station and other infrastructure projects. SUMMARY

[0004] The present application aims to solve the problem that the existing technology uses traditional empirical icing models and mechanism analysis methods, and does not apply assimilated meteorological prediction data to short-term icing prediction and early warning of power lines. In addition, due to the remote location of power lines, there is a lack of long-term icing observation and fine meteorological observation of multi-source data, which cannot provide data that meet the spatial and temporal accuracy and prediction accuracy to meet the demand for predicting the thickness of power line icing.

[0005] The present application proposes a power line icing simulation prediction method based on data assimilation and artificial intelligence, which fully utilizes long-term continuous observation data, conducts research on advanced multi-source meteorological data assimilation methods, and outputs fine meteorological prediction information field. The artificial intelligence integrated constraint model is used to correct the meteorological prediction data field, obtain the corrected meteorological information, calculate the spatial distribution field of icing by mathematical method, and output the icing thickness prediction result, which can assist in solving the power line icing early warning problem.

[0006] The present application comprises the following steps:

[0007] Step 1, based on a meteorological mesoscale numerical model, the meteorological elements affecting the generation of power line icing are predicted; the meteorological elements include temperature, humidity, wind speed and precipitation;

[0008] Step 2, based on the regular meteorological observation data, the observation data of the automatic station on the high mountain and the icing data, the GSI three-dimensional variation system is used to carry out the initial field analysis of the wire icing data, the assimilation of the meteorological elements related to the generation of the wire icing is realized, the background element field is adjusted according to the following formula to obtain a more accurate initial field, and thus the result of the numerical mode prediction in step 1 is optimized:

[0009] Step 3, the result of the numerical mode prediction optimized in step 2 is subjected to bias correction processing through the artificial intelligence integrated nested model based on different machine learning;

[0010] Step 4, based on the meteorological elements subjected to the bias correction processing obtained in step 3, it is determined whether the icing is generated, the geographical position where the wire icing is likely to occur is marked, and the icing spatial distribution information field of the concerned area is formed;

[0011] Step 5, according to the icing thickness growth mathematical formula, the thickness of the icing is calculated for the icing spatial distribution information field of the concerned area obtained in step 4, and the warning result of the regional icing thickness is output.

[0012] In step 1, the Weather Research and Forecasting Model 4.4 is selected as the meteorological mesoscale numerical mode, which is the latest updated meteorological prediction model, and this version is first applied to the prediction of the meteorological elements related to the icing prediction. In the prediction process, the global prediction is selected as the mode driving field, the Thompson microphysical scheme and the YSU (Yonsei University) boundary layer scheme are combined to predict the meteorological elements in the target area, and the prediction results of the meteorological elements including temperature, humidity, wind speed and precipitation are provided for the icing state determination and the simulation prediction of the icing thickness.

[0013] In step 2, the GSI three-dimensional variation system is used to carry out the initial field analysis of the wire icing data in combination with the regular meteorological observation data, the observation data of the automatic station on the high mountain and the icing data, the assimilation of the meteorological elements related to the generation of the wire icing is realized, the background element field is adjusted according to the following formula to obtain a more accurate initial field, and thus the result of the numerical mode prediction in step 1 is optimized:

[0014]

[0015] Wherein J(x) represents the minimized objective function (functional), x represents the analysis field, x b represents the background field, H(x) is the observation operator, y is the observation value, B and R are the background error covariance matrix and the observation error covariance matrix respectively, represents the relative contribution to the analysis field; T represents the matrix transposition.

[0016] The icing station history and real-time data are relatively difficult to obtain, the application adds an icing data interface in the GSI system, and the icing data is used for assimilation for the first time; meanwhile, the high mountain station and the meteorological conventional data are simultaneously assimilated into the GSI system, the meteorological element prediction results of wind speed, temperature, relative humidity and precipitation are optimized, and the icing simulation is further optimized.

[0017] In step 3, integrated nested models are established for different targets: temperature, humidity, wind speed and precipitation, the integrated nested models have multi-output functions and include a basic data processing framework. In the model training process, the models of different targets can be independently trained, and the parameter factors and weight coefficients of each model are obtained through the learning of the historical data law. Since the meteorological factors have complex physical correlations with each other, in the model nesting selection, the correlation information of the multi-output factors of wind speed, temperature and precipitation can be considered.

[0018] In step 3, the basic data processing framework includes a streamlined processing module integrating intelligent quality control models, data normalization and data standardization;

[0019] The basic data processing framework is used to obtain temperature, humidity, wind speed and precipitation data of numerical mode prediction and real-time observation;

[0020] The intelligent quality control model fuses models for temperature, wind speed, relative humidity and precipitation to intelligently identify and determine data, for temperature data, a periodic change model of historical contemporaneous temperature is used for abnormal detection and periodic detection of daily temperature change and monthly temperature change; for relative humidity and precipitation data, a historical statistical periodic detection model is used for trend and consistency detection; for wind speed data, abnormal value identification and detection are performed to remove data that does not meet the quality control requirements, and the remaining data is input into the streamlined processing module integrating data normalization and data standardization, and normalization and standardization calculation are performed according to the following formula:

[0021] The normalization formula is:

[0022]

[0023] Where x0 represents the original value, x min represents the minimum value, x max represents the maximum value, and x1 represents the normalized value.

[0024] The standardization formula is:

[0025] x2=(x0-μ) / σ,

[0026] Where x2 represents the standardized data, and sigma represents the standard deviation of the original data.

[0027] After normalization and standardization, output feature data is generated, which is matched with the observed data.

[0028] In step 3, the ensemble nested model utilizes CatBoost (a type of gradient boosting algorithm), GBDT (Gradient Boosting Decision Tree), AdaBoost (Adaptive Boosting Decision Tree), and RandomForest (Random Forest) models, combining Boosting and Bagging approaches for ensemble generation. The entire ensemble nested model consists of three layers. CatBoost is located at the bottom layer (the first layer). Feature data obtained from the basic data processing framework enters CatBoost, and the output of the training with fixed parameters is used as input features. This output is then sent to the GBDT and AdaBoost models in the second layer for training. Again, the training results with fixed parameters are then superimposed on the RandomForest model in the third layer. Multi-threaded, multi-module distributed computation is implemented, and modeling is performed separately for temperature, humidity, wind speed, and precipitation. In the second layer of the ensemble nested model, the training results of the AdaBoost and GBDT models are used in a Bagging manner, forming the nested result of the second layer according to the following formula:

[0029]

[0030] Where F2 refers to the nested result of the second layer, f(AdaBoost) refers to the training result of the AdaBoost model, and f(GBDT) refers to the training result of the GBDT gradient boosting decision tree.

[0031] The final ensemble nested model output uses a boosting method on the three-layer model results, and the corrected final output is formed according to the following formula:

[0032] F3=0.2*f(CatBoost)+0.5*F2(AdaBoost+GBDT)+0.3*f1,

[0033] Where F3 refers to the nested result of the third layer, f(CatBoost) refers to the training result of the underlying CatBoost model, f1 refers to the training result of the RandomForest model, and F2(AdaBoost+GBDT) refers to the training result of the second layer model.

[0034] Based on the differences between input features and real-world observation labels, the parameters of different machine learning models are updated and adjusted during the modeling process, and training continues until the training stop condition is met. The trained ensemble nested model is then used to perform bias correction processing on temperature, humidity, wind speed, and precipitation.

[0035] In step 3, the update and adjustment of different machine learning model parameters includes adjusting the number of weak learners, the number of trees, and the depth of the trees. Different njobs threads are set for different factors, namely wind speed, humidity, temperature, and precipitation. Only one factor is focused on in a single thread. During the adjustment of the ensemble nested model, the layer order of CatBoost, GBDT gradient boosting decision tree, and RandomForest is adjusted sequentially until the training meets the requirements.

[0036] In step 4, when the precipitation in the input meteorological elements is 0, and the temperature T≤1℃, relative humidity≥90%, and wind speed SPD≤6m / s are satisfied, it is determined that icing has occurred; otherwise, it is determined that icing has not occurred, thus forming an information field of icing spatial distribution in the area of ​​interest.

[0037] In step 5, combining the spatial distribution information field of icing in the area of ​​interest, the icing thickness at locations where icing is likely to occur is calculated using the following formula:

[0038] in The value represents the mass of ice growth on the transmission line per unit time, where d represents the differential, ν represents the velocity of the incident particle (replaced by wind speed), W represents the liquid water content, α1 represents the collision rate, α2 represents the capture rate, and α3 represents the freezing coefficient. When icing is determined to occur, the thickness A of the icing is calculated according to the formula, which is the effective cross-sectional area of ​​the object.

[0039] The present invention also provides a storage medium storing a computer program or instructions, which, when the computer program or instructions are run, implement the aforementioned method for simulating and forecasting power line icing based on data assimilation and artificial intelligence.

[0040] Beneficial effects: Compared with the prior art, the advantages of this invention are:

[0041] This invention utilizes the latest versions of WRF and GSI, and for the first time considers meteorological numerical model forecast assimilation in the power line icing thickness prediction model.

[0042] This invention integrates years of observation data, making full use of long-term continuous icing observation data and data from high-altitude stations to conduct cutting-edge numerical assimilation model research.

[0043] This invention proposes an integrated artificial intelligence framework that integrates multiple single AI models to constrain and correct various meteorological elements predicted by numerical models, thereby obtaining more accurate meteorological element data. The meteorological element data is then combined with mathematical empirical models to predict icing thickness. Attached Figure Description

[0044] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, and the advantages of the present invention in the above and / or other aspects will become clearer.

[0045] Figure 1 This is a flowchart of the method of the present invention.

[0046] Figure 2 This is a schematic diagram of the basic data processing framework of the present invention.

[0047] Figure 3 This is a schematic diagram of the integrated nested model of the present invention.

[0048] Figure 4 This is a schematic diagram of the temporal distribution of meteorological elements in numerical model assimilation forecasts. Detailed Implementation

[0049] This invention provides a method for simulating and forecasting power line icing based on data assimilation and artificial intelligence, comprising the following steps:

[0050] Step 1: Based on a meteorological mesoscale numerical model, forecast the meteorological elements that affect the formation of icing on transmission lines; the meteorological elements include temperature, humidity, wind speed, and precipitation.

[0051] Step 2: Combining regional routine meteorological observations, observation data from high-altitude automatic weather stations, and icing data, the three-dimensional variational assimilation system Gridpoint Statistical Interpolation is used to perform assimilation analysis of icing station data, adjust the initial field of the meteorological model forecast, and optimize the results of the numerical model forecast in Step 1.

[0052] Step 3: By integrating nested artificial intelligence models based on different machine learning methods, bias correction is performed on the numerical model forecast results optimized in Step 2.

[0053] Step 4: Based on the meteorological elements after deviation correction obtained in Step 3, determine whether icing has occurred, mark the geographical locations where power line icing may occur, and form a spatial distribution information field of icing in the area of ​​interest.

[0054] Step 5: For the icing spatial distribution information field of the area of ​​interest obtained in Step 4, calculate the icing thickness according to the mathematical formula for icing thickness growth, and output the early warning result of the icing thickness of the area.

[0055] In step 1, the meteorological mesoscale numerical model used is Weather Research and Forecasting Model 4.4, which is the latest updated meteorological forecasting model. This version is being used for the first time to forecast meteorological elements related to icing prediction. In the forecast processing, global forecasts are selected as the model driving field, and the Thompson microphysics scheme and the YSU (Yonsei University) boundary layer scheme are combined to forecast meteorological elements in the target area, providing forecast results of meteorological elements including temperature, humidity, wind speed and precipitation for icing state determination and icing thickness simulation prediction.

[0056] In step 2, combining regional routine meteorological observations, observation data from high-altitude automatic weather stations, and icing data, the GSI three-dimensional variational system is used to conduct initial field analysis of power line icing data. This assimilates meteorological elements related to power line icing formation, and adjusts the background element field according to the following formula to obtain a more accurate initial field, thereby optimizing the results of the numerical model forecast in step 1:

[0057]

[0058] Where J(x) represents the minimization objective function (functional), x represents the analysis field, and x b H(x) represents the background field, y is the observation operator, B and R are the background error covariance matrix and the observation error covariance matrix, respectively, representing their relative contributions to the analysis field; T represents the matrix transpose.

[0059] Historical and real-time data from icing stations are relatively difficult to obtain. This invention adds an interface for icing data to the GSI system and uses icing data for assimilation for the first time. At the same time, high-altitude station and routine meteorological data are assimilated into the GSI system to optimize the forecast results of meteorological elements such as wind speed, temperature, relative humidity and precipitation, and further optimize icing simulation.

[0060] In step 3, ensemble nested models are established for different objectives: temperature, humidity, wind speed, and precipitation. These models have multi-output capabilities and include a basic data processing framework. During model training, models for different objectives can be trained independently, and the parameter factors and weight coefficients of each model are obtained by learning from historical data patterns. Since meteorological factors have complex physical correlations, the correlation information of multi-output factors such as wind speed, temperature, and precipitation can be considered when selecting nested models.

[0061] In step 3, the basic data processing framework includes a streamlined processing module that integrates intelligent quality control model, data normalization and data standardization.

[0062] The basic data processing framework is used to acquire temperature, humidity, wind speed and precipitation data from numerical model forecasts and real-world observations;

[0063] The integrated intelligent quality control model constructs models for temperature, wind speed, relative humidity, and precipitation to intelligently identify and judge data. For temperature data, it uses a periodic variation model based on historical temperature data to detect anomalies and cycles in daily and monthly temperature variations. For relative humidity and precipitation data, it uses a periodic detection model based on historical statistics to detect trends and consistency. For wind speed data, it identifies and detects outliers, removes data that does not meet quality control requirements, and inputs the remaining data into a streamlined processing module that integrates data normalization and standardization. The normalization and standardization calculations are then performed according to the following formula.

[0064] The normalization formula is:

[0065]

[0066] Where x0 represents the original value, x min x represents the minimum value. max x1 represents the maximum value, and x2 represents the normalized value.

[0067] The standardized formula is:

[0068] x2=(x0-μ) / σ,

[0069] Where x2 represents the standardized data, and σ represents the standard deviation of the original data;

[0070] After normalization and standardization, output feature data is generated, which is matched with the observed data.

[0071] In step 3, the ensemble nested model utilizes CatBoost (a type of gradient boosting algorithm), GBDT (Gradient Boosting Decision Tree), AdaBoost (Adaptive Boosting Decision Tree), and RandomForest (Random Forest) models, combining Boosting and Bagging approaches for ensemble generation. The entire ensemble nested model consists of three layers. CatBoost is located at the bottom layer (the first layer). Feature data obtained from the basic data processing framework enters CatBoost, and the output of the training with fixed parameters is used as input features. This output is then sent to the GBDT and AdaBoost models in the second layer for training. Again, the training results with fixed parameters are then superimposed on the RandomForest model in the third layer. Multi-threaded, multi-module distributed computation is implemented, and modeling is performed separately for temperature, humidity, wind speed, and precipitation. In the second layer of the ensemble nested model, the training results of the AdaBoost and GBDT models are used in a Bagging manner, forming the nested result of the second layer according to the following formula:

[0072]

[0073] Where F2 refers to the nested result of the second layer, f(AdaBoost) refers to the training result of the AdaBoost model, and f(GBDT) refers to the training result of the GBDT gradient boosting decision tree.

[0074] The final ensemble nested model output uses a boosting method on the three-layer model results, and the corrected final output is formed according to the following formula:

[0075] F3=0.2*f(CatBoost)+0.5*F2(AdaBoost+GBDT)+0.3*f1,

[0076] Where F3 refers to the nested result of the third layer, f(CatBoost) refers to the training result of the underlying CatBoost model, f1 refers to the training result of the RandomForest model, and F2(AdaBoost+GBDT) refers to the training result of the second layer model.

[0077] Based on the differences between input features and real-world observation labels, the parameters of different machine learning models are updated and adjusted during the modeling process, and training continues until the training stop condition is met. The trained ensemble nested model is then used to perform bias correction processing on temperature, humidity, wind speed, and precipitation.

[0078] In step 3, the update and adjustment of different machine learning model parameters includes adjusting the number of weak learners, the number of trees, and the depth of the trees. Different njobs threads are set for different factors, namely wind speed, humidity, temperature, and precipitation. Only one factor is focused on in a single thread. During the adjustment of the ensemble nested model, the layer order of CatBoost, GBDT gradient boosting decision tree, and RandomForest is adjusted sequentially until the training meets the requirements.

[0079] In step 4, when the precipitation in the input meteorological elements is 0, and the temperature T≤1℃, relative humidity≥90%, and wind speed SPD≤6m / s are satisfied, it is determined that icing has occurred; otherwise, it is determined that icing has not occurred, thus forming an information field of icing spatial distribution in the area of ​​interest.

[0080] In step 5, combining the spatial distribution information field of icing in the area of ​​interest, the icing thickness at locations where icing is likely to occur is calculated using the following formula:

[0081] in The value represents the mass of ice growth on the transmission line per unit time, where d represents the differential, ν represents the velocity of the incident particle (replaced by wind speed), W represents the liquid water content, α1 represents the collision rate, α2 represents the capture rate, and α3 represents the freezing coefficient. When icing is determined to occur, the thickness A of the icing is calculated according to the formula, which is the effective cross-sectional area of ​​the object.

[0082] The present invention also provides a storage medium storing a computer program or instructions, which, when the computer program or instructions are run, implement the aforementioned method for simulating and forecasting power line icing based on data assimilation and artificial intelligence.

[0083] Example

[0084] Various aspects of the invention are described in this disclosure with reference to the accompanying drawings, which illustrate numerous illustrative embodiments. The embodiments of this disclosure are not necessarily intended to encompass all aspects of the invention. It should be understood that the various concepts and embodiments described above, as well as those described in more detail below, can be implemented in any of many ways, because the concepts and implementations disclosed herein are not limited to any particular implementation. Furthermore, some aspects of the invention disclosed may be used alone or in any suitable combination with other aspects of the invention disclosed.

[0085] This embodiment proposes a method for simulating and forecasting power line icing based on data assimilation and artificial intelligence. It utilizes the latest numerical model forecast assimilation technology to output forecast fields of different meteorological elements. An integrated multi-model artificial intelligence method is then used to correct the assimilated forecast fields, obtaining more accurate information on temperature, humidity, precipitation, and wind speed. Based on the aforementioned meteorological element forecast results, the spatial distribution field of power line icing is obtained, and the predicted icing thickness in the icing-affected areas is output using mathematical empirical formulas.

[0086] Figure 1 This invention presents the flowchart of a power line icing simulation and forecasting method based on data assimilation and artificial intelligence. The invention combines meteorological element assimilation forecast correction with mathematical empirical formulas. Figure 1 The process shown includes:

[0087] Step 1 is based on the latest version of the meteorological mesoscale numerical model Weather Research and Forecasting Model 4.4 (WRF v4.4), and for the first time forecasts meteorological elements such as temperature, humidity, wind speed and precipitation that affect the formation of icing on transmission lines;

[0088] The aforementioned mesoscale meteorological numerical model uses WRF v4.4 to simulate meteorological elements within the target area, providing meteorological information such as temperature, humidity, wind, and precipitation. The specific parameter settings for the WRF model are shown in Table 1. The simulation center latitude and longitude are 103°E and 29°E; the horizontal grid number is set to 352*302, and the model resolution is 3km.

[0089] Table 1

[0090]

[0091] Step 2 combines routine regional meteorological observations and observation data from high-altitude automatic weather stations. For the first time, the latest version of the three-dimensional variational assimilation system Gridpoint Statistical Interpolation (GSI) is used to perform assimilation analysis on icing station data, adjust the initial field of meteorological model forecasts, and optimize the results of numerical model forecasts.

[0092] The assimilated data includes regional routine meteorological observations and observations from high-altitude automatic weather stations. The GSI three-dimensional variational (3DVAR) assimilation (DA) system was used to effectively assimilate meteorological elements from the stations, according to the specific formula: Adjusting the background field provides a more reasonable and accurate initial field for later model forecasts.

[0093] Step 3 adopts an integrated nested model ( Figure 3The method corrects biases in the basic meteorological elements such as temperature, humidity, and wind output from numerical model forecasts; the basic data processing framework ( Figure 2 It encompasses a series of data mining modules, including basic data input, data analysis and processing, normalization and standardization, with input and output in a pipeline format to achieve preprocessing of various types of data.

[0094] For icing-related meteorological factors—temperature, humidity, wind speed, and precipitation—separate training models are established. Based on these, a nested model is integrated, incorporating multiple machine learning models. Each target can be accurately represented by a single model, and information about the target is obtained by examining the corresponding regressor. During model training, models for different targets can be trained independently, and the parameter factors and weight coefficients of each model are obtained by learning from historical data patterns. A nested strength-adaptive AdaBoost and Random Forest models are integrated to obtain information about the target. Specifically, the `n_estimators` value of the RandomForestRegressor is set to 1000, and the `max_features` value is set to 0.75 to enhance the prediction effect for different factors. Meanwhile, the `n_estimators` value of AdaBoost is set to 50.

[0095] Because meteorological factors have complex physical correlations with each other, multiple output factors, namely the correlation information of wind speed, temperature and precipitation, are considered in the nested selection of models.

[0096] Step 4, based on the meteorological elements output in Step 3, uses empirical formulas to determine whether power line icing has occurred. Geographic locations where power line icing is likely to occur are marked, ultimately forming the spatial distribution of icing. Icing is determined to have occurred when the input meteorological elements are precipitation 0, and the conditions are met: temperature T ≤ 1℃, relative humidity ≥ 90%, and wind speed SPD ≤ 6m / s; otherwise, icing does not occur. For an icing event in December 2020, the Gaoqiao station was selected, and the occurrence of icing was determined using wind speed, temperature, and relative humidity predicted by the numerical model. Figure 4 It is a time series diagram of the model forecast, selecting the time period that meets the meteorological conditions for icing.

[0097] Step 5 calculates the icing thickness at the locations identified in Step 4 based on the mathematical formula for icing thickness growth, and outputs the early warning result for the icing thickness of the area.

[0098] For sites meeting the conditions for icing, icing thickness is predicted using a mathematical empirical formula. The calculation formula is as follows: in This represents the mass increase in icing on the transmission line per unit time. When icing is detected, the icing thickness is calculated using the formula: A is the effective cross-sectional area of ​​the object, ν represents the velocity of the incident particle (represented by wind speed), W represents the liquid water content, α1 represents the collision rate, α2 represents the capture rate, and α3 represents the freezing coefficient. In this example, the icing thickness on the power line at the location where icing occurs is calculated.

[0099] This embodiment proposes a method for simulating and forecasting power line icing based on meteorological numerical forecast assimilation and artificial intelligence integrated constraint correction. It utilizes numerical model forecast assimilation technology to output forecast fields of different meteorological elements, and then corrects these fields using an artificial intelligence method integrating multiple models to obtain more accurate information on temperature, humidity, and wind speed. Based on the aforementioned meteorological element forecast results, the spatial distribution field of power line icing is obtained, and the predicted icing thickness in the icing-affected areas is output using mathematical empirical formulas, enabling effective early warning of power line icing.

[0100] This invention provides a method for simulating and forecasting power line icing based on data assimilation and artificial intelligence. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment of the invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

Claims

1. A method for simulating and forecasting power line icing based on data assimilation and artificial intelligence, characterized in that, Includes the following steps: Step 1: Based on a meteorological mesoscale numerical model, forecast the meteorological elements that affect the formation of icing on transmission lines; the meteorological elements include temperature, humidity, wind speed, and precipitation. Step 2: Combine regional routine meteorological observations, observation data from high-altitude automatic weather stations, and icing data to conduct assimilation analysis of icing station data, adjust the initial field of the meteorological model forecast, and optimize the results of the numerical model forecast in Step 1. Step 3: By integrating nested artificial intelligence models based on different machine learning methods, bias correction is performed on the numerical model forecast results optimized in Step 2. Step 4: Based on the meteorological elements after deviation correction obtained in Step 3, determine whether icing has occurred, mark the geographical locations where power line icing may occur, and form a spatial distribution information field of icing in the area of ​​interest. Step 5: For the spatial distribution information field of ice accretion in the area of ​​interest obtained in Step 4, calculate the ice thickness according to the mathematical formula for ice thickness growth, and output the early warning result of the ice thickness in the area. In step 3, the ensemble nested model utilizes CatBoost, GBDT gradient boosting decision tree, adaptive boosting AdaBoost model, and RandomForest model, combining boosting and bagging approaches for ensemble generation. The entire ensemble nested model consists of three layers, with CatBoost at the bottom layer (the first layer). Feature data obtained from the basic data processing framework enters CatBoost, and the output of the training with fixed parameters is used as input features. This output is then sent to the GBDT gradient boosting decision tree and the adaptive boosting AdaBoost model in the second layer for training. The training results with fixed parameters are then superimposed on the RandomForest model in the third layer. Multi-threaded, multi-module distributed computation is implemented, and modeling is performed separately for temperature, humidity, wind speed, and precipitation. In the second layer of the ensemble nested model, the training results of the adaptive boosting AdaBoost model and GBDT gradient boosting decision tree are used in a bagging manner, forming the nested result of the second layer according to the following formula: Where F2 refers to the nested result of the second layer, f(AdaBoost) refers to the training result of the AdaBoost model, and f(GBDT) refers to the training result of the GBDT gradient boosting decision tree; The final ensemble nested model output uses a boosting method on the three-layer model results, and the corrected final output is formed according to the following formula: F3=0.2*f(CatBoost)+0.5*F2(AdaBoost+GBDT)+0.3*f1, Where F3 refers to the nested result of the third layer, f(CatBoost) refers to the training result of the underlying CatBoost model, f1 refers to the training result of the RandomForest model, and F2(AdaBoost+GBDT) refers to the training result of the second layer model. Based on the differences between input features and real-world observation labels, the parameters of different machine learning models are updated and adjusted during the modeling process, and training continues until the training stop condition is met. The trained ensemble nested model is then used to perform bias correction processing on temperature, humidity, wind speed, and precipitation.

2. The method according to claim 1, characterized in that, In step 1, the meteorological mesoscale numerical model used is Weather Research and Forecasting Model 4.

4. During the forecast processing, global forecasts are selected as the model driving field. The Thompson microphysics scheme and the YSU boundary layer scheme are combined to forecast meteorological elements in the target area, providing forecast results of meteorological elements including temperature, humidity, wind speed and precipitation for icing status determination and icing thickness simulation prediction.

3. The method according to claim 2, characterized in that, In step 2, combining regional routine meteorological observations, observation data from high-altitude automatic weather stations, and icing data, the GSI three-dimensional variational system is used to conduct initial field analysis of power line icing data. This assimilates meteorological elements related to power line icing formation, and adjusts the background element field according to the following formula to obtain a more accurate initial field, thereby optimizing the results of the numerical model forecast in step 1: Where J(x) represents the minimization objective function, x represents the analysis field, and x b H(x) represents the background field, y is the observation operator, B and R are the background error covariance matrix and the observation error covariance matrix, respectively, representing their relative contributions to the analysis field; T represents the matrix transpose.

4. The method according to claim 3, characterized in that, In step 3, an integrated nested model is established for different targets: temperature, humidity, wind speed, and precipitation. The integrated nested model has multiple output functions and includes a basic data processing framework.

5. The method according to claim 4, characterized in that, In step 3, the basic data processing framework includes a streamlined processing module that integrates intelligent quality control model, data normalization and data standardization. The basic data processing framework is used to acquire temperature, humidity, wind speed and precipitation data from numerical model forecasts and real-world observations; The integrated intelligent quality control model constructs models for temperature, wind speed, relative humidity, and precipitation to intelligently identify and judge data. For temperature data, it uses a periodic variation model based on historical temperature data to detect anomalies and cycles in daily and monthly temperature variations. For relative humidity and precipitation data, it uses a periodic detection model based on historical statistics to detect trends and consistency. For wind speed data, it identifies and detects outliers, removes data that does not meet quality control requirements, and inputs the remaining data into a streamlined processing module that integrates data normalization and standardization. The normalization and standardization calculations are then performed according to the following formula. The normalization formula is: Where x0 represents the original value, x min x represents the minimum value. max x1 represents the maximum value, and x2 represents the normalized value. The standardized formula is: x2=(x0-μ) / σ, Where x2 represents the standardized data, and σ represents the standard deviation of the original data; After normalization and standardization, output feature data is generated, which is matched with the observed data.

6. The method according to claim 5, characterized in that, In step 3, the update and adjustment of different machine learning model parameters includes adjusting the number of weak learners, the number of trees, and the depth of the trees. Different njobs threads are set for different factors, namely wind speed, humidity, temperature, and precipitation. Only one factor is focused on in a single thread. During the adjustment of the ensemble nested model, the layer order of CatBoost, GBDT gradient boosting decision tree, and RandomForest is adjusted sequentially until the training meets the requirements.

7. The method according to claim 6, characterized in that, In step 4, when the precipitation in the input meteorological elements is 0, and the temperature T≤1℃, relative humidity≥90%, and wind speed SPD≤6m / s are satisfied, it is determined that icing has occurred; otherwise, it is determined that icing has not occurred, thus forming an information field of icing spatial distribution in the area of ​​interest.

8. The method according to claim 7, characterized in that, In step 5, combining the spatial distribution information field of icing in the area of ​​interest, the icing thickness at locations where icing is likely to occur is calculated using the following formula: in The value represents the mass of ice growth on the transmission line per unit time, where d represents the differential, ν represents the velocity of the incident particle, W represents the liquid water content, α1 represents the collision rate, α2 represents the capture rate, and α3 represents the freezing coefficient. When icing is determined to have occurred, the thickness A of the icing is calculated using the formula and is the effective cross-sectional area of ​​the object.

9. A storage medium, characterized in that, It stores a computer program or instructions that, when executed, implement the method as described in any one of claims 1 to 8.

Images