Method for forecasting icing thickness probability of power transmission line in complex terrain based on multi-mode set

Abstract

The invention discloses a complex terrain power transmission line icing thickness probability forecasting method based on a multi-mode set, and relates to the technical field of disaster prevention and reduction and weather forecasting of a power system. Constructing a plurality of icing thickness prediction models based on a physical mechanism and artificial intelligence based on the forecast data; and integrating the icing thickness prediction models into one icing thickness probability prediction method, and quantifying the uncertainty of icing prediction. Through the method, the accuracy and robustness of icing thickness prediction under complex terrain conditions can be effectively improved, a scientific basis with certainty and probabilistic is provided for power grid ice prevention and disaster reduction decisions, and thus the risk early warning and prevention and control capability of a power grid to deal with ice disasters is enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system disaster prevention and mitigation and meteorological forecasting, and particularly relates to a complex terrain transmission line icing thickness probability forecasting method based on a multi-mode set. BACKGROUND

[0002] Transmission line icing is one of the natural disasters that seriously threatens the safe operation of the power grid. Accurate prediction of the icing thickness and its evolution trend of the transmission line is of great significance for carrying out anti-icing and de-icing work and ensuring the stable operation of the power grid.

[0003] Existing icing prediction methods can be mainly divided into two categories according to their principles: The first category is a prediction model based on physical mechanisms. This type of model describes the influence relationship of icing formation through mathematical expressions based on the principles of fluid mechanics and thermodynamics. Among them, the Imai model is suitable for wet growth icing; the Goodwin model and the Chain model are mainly for dry growth icing; the Lenhard model mainly considers the influence of precipitation; the Jones model focuses on the analysis of precipitation and wind speed and is suitable for prediction of glaze icing; the Makkonen model integrates fluid mechanics, thermodynamics and meteorology, and introduces the concepts of capture rate, freezing rate and collision rate to build a prediction model for rime icing. The shortcomings of this type of model are: the physical process of icing formation is extremely complex, the existing models have limitations and pertinence in describing the physical mechanism, and there are different degrees of simplification in the mathematical modeling process, thereby restricting the improvement of prediction accuracy.

[0004] The second category is a prediction model based on artificial intelligence. This type of model does not rely on the explanation of physical mechanisms. From the strategic principle, it can be divided into time series machine learning algorithm and nonlinear regression machine learning method. The former can learn the time sequence relationship between historical meteorological factors, historical icing thickness and the icing thickness to be predicted, and the latter can capture the complex nonlinear mapping between related factors such as meteorological elements and icing formation. The shortcomings of this type of model are: the model training highly depends on the accuracy and refinement of meteorological prediction data, especially in complex terrain areas, the local nature of meteorological elements is strong, resulting in large icing prediction error.

[0005] In the prior art, for icing thickness prediction, Chinese patents CN120087747A and CN112711919A combine mesoscale weather models with physical mechanism models, while Chinese patent CN107092982A adopts an artificial intelligence prediction model. Their common limitations are: models relying on a single principle are difficult to guarantee robustness and accuracy when facing complex and variable icing processes; in addition, existing methods usually only output a certain icing thickness value or icing level, and cannot provide credibility evaluation of the prediction results.

[0006] It can be seen that the defects of the existing power transmission line icing prediction technology mainly lie in three aspects: (1) The fine prediction of meteorological elements under complex terrain conditions has significant deviation. Complex terrain can cause local climate phenomena, leading to high heterogeneity of key meteorological parameters on a microscale. Existing mesoscale meteorological models are limited by resolution and parameterization schemes, making it difficult to accurately depict the hundred-meter or even ten-meter meteorological element gradient caused by terrain dynamic and thermal effects. This "systematic deviation" of meteorological input directly restricts the accuracy of subsequent icing prediction, especially for data-driven artificial intelligence models, making the micro-terrain area a blind spot for icing risk prevention and control.

[0007] (2) Single modeling principle cannot fully represent the complex process of icing which involves the nonlinear interaction of multiple physical processes such as aerodynamics, thermodynamics, and phase change mass transfer. The complexity of icing formation far exceeds the representation ability of a single model. Although the mechanism-based physical model has a clear physical basis, it often fails due to excessive simplification when facing abnormal icing events beyond the model's assumption conditions; while the data-driven artificial intelligence model has strong nonlinear fitting ability, but its performance is strictly constrained by the completeness of training data, and the prediction reliability under extreme conditions is questionable. The two types of models have not formed effective collaboration at the methodological level, restricting the robustness and generalization ability of the prediction system.

[0008] (3) The deterministic prediction paradigm cannot support risk warning-based disaster prevention decision-making. Existing prediction systems output single-point estimates or discrete levels, failing to quantify the uncertainty of the prediction results themselves. In power grid disaster prevention decision-making, not only the point prediction of icing thickness is needed, but also the probability distribution characteristics and the confidence level of the prediction results. The current lack of probability prediction and risk quantification tools makes it difficult to achieve the optimal trade-off between prevention costs and failure losses, and to make more scientific and forward-looking risk warning and disaster prevention resource allocation. SUMMARY

[0009] The present application overcomes the above technical deficiencies and provides a power transmission line icing prediction method combining meteorological numerical simulation, machine learning and ensemble prediction technology, especially suitable for fine icing probability prediction under complex terrain.

[0010] The technical solution adopted by the present application to overcome the technical problems is: A complex terrain power transmission line icing thickness probability prediction method based on multi-model ensemble, comprising: S1. Obtain a power transmission line icing historical data set ; S2. Obtain a power transmission line icing historical data set The meteorological simulation data of the same period is obtained as a meteorological simulation data set ; S3. Using the meteorological simulation data set and the historical icing data set of the power transmission line establishing an icing prediction machine learning model, and obtaining a historical icing thickness prediction sequence of the power transmission line according to the icing prediction machine learning model ; S4. Constructing an icing prediction mechanism model, according to the meteorological simulation data set , the historical icing data set of the power transmission line and the icing prediction mechanism model to obtain a historical icing thickness prediction sequence of the power transmission line ; S5. Using the historical icing thickness prediction sequence of the power transmission line , the historical icing thickness prediction sequence of the power transmission line and the historical icing data set of the power transmission line establishing an icing set probability prediction machine learning model, and training the icing set probability prediction machine learning model to obtain an optimized icing set probability prediction machine learning model S6. Constructing a meteorological prediction data set of a future period , the meteorological prediction data set is input into the optimized icing prediction machine learning model, the icing prediction mechanism model, and the optimized icing set probability prediction machine learning model, and the output is obtained as an icing thickness probability prediction result of the future period.

[0011] Further, in step S1, 5 years of icing monitoring data is collected from a certain area to obtain an icing monitoring data set , the icing monitoring data including time s, power transmission line name id, power transmission line monitoring icing thickness h, power transmission line conductor diameter d0, and power transmission line vector information shp.

[0012] Further, step S2 includes the following steps: S2-1. Using meteorological reanalysis data to drive a mesoscale weather model WRF to generate a mesoscale meteorological field, and using a CALWRF preprocessor to process the mesoscale meteorological field into meteorological data 3D.DAT readable by a microscale diagnostic model CALMET S2-2. Collect the geophysical data required by the microscale diagnostic model CALMET, including SRTMDEM 90m resolution DEM terrain elevation data, ESRI Land Cover 10m resolution land use data, and sequentially using TERREL preprocessor, CTGPROC preprocessor and MAKEGEO preprocessor to process the geophysical data into the geophysical data GEO.DAT readable by the microscale diagnostic model CALMET; S2-3. For the target transmission line area, set the parameterization scheme in the microscale diagnostic model CALMET to turn off terrain dynamics, turn on O'Brien vertical velocity, turn on downslope airflow effect, turn off Froude number, output time resolution is 1h, output horizontal spatial resolution is 50m, vertical height is 10m, get small microscale wind field prediction model WRF-CALMET; S2-4. Input the weather data 3D.DAT as the initial guess field and the geophysical data GEO.DAT as the environment field into the small microscale wind field prediction model WRF-CALMET, sequentially perform terrain dynamics, slope flow, terrain blocking effect, radiation minimization adjustment, interpolation smoothing, vertical velocity calculation, generate weather gridded simulation data, and use the same time s, same transmission line name id, and same transmission line vector information shp to extract the meridional wind u, zonal wind v, air temperature t, air pressure p, precipitation q, freezing rain z, and relative humidity r at the transmission line from the weather gridded simulation data, and the weather simulation data set including time s, transmission line name id, meridional wind u, zonal wind v, air temperature t, air pressure p, precipitation q, freezing rain z, and relative humidity r. including time s, transmission line name id, meridional wind u, zonal wind v, air temperature t, air pressure p, precipitation q, freezing rain z, and relative humidity r.

[0013] Further, step S3 includes the following steps: S3-1. The icing prediction machine learning model includes a first type of model and a second type of model, a number of models a are selected from the first type of model, and a number of models b are selected from the second type of model, a is greater than or equal to 1, b is greater than or equal to 1, the first type of model includes long short-term memory network and gated recurrent unit; the second type of model includes support vector regression and random forest; S3-2. The weather simulation data set is divided into weather simulation training set M1, weather simulation verification set M2 and weather simulation test set M3 in the time ratio of 4:1:5, and the transmission line icing thickness h in the transmission line icing history data set is divided into icing thickness training set R1, icing thickness verification set R2 and icing thickness test set R3 in the time ratio of 4:1:5; S3-3. inputting the meteorological simulation training set M1 as input features and the icing thickness training set R1 as training targets into the first type of model and the second type of model respectively to obtain a relationship model of the first type of model and a relationship model of the second type of model; S3-4. using an optimization algorithm to use the meteorological simulation verification set M2 as input features and the icing thickness verification set R2 as verification targets to independently optimize the hyperparameters of the relationship model of the first type of model, save the best parameter configuration of the relationship model of the first type of model, obtain the trained first type of model, and use the optimization algorithm to use the meteorological simulation verification set M2 as input features and the icing thickness verification set R2 as verification targets to independently optimize the hyperparameters of the relationship model of the second type of model, save the best parameter configuration of the relationship model of the second type of model, and obtain the trained second type of model; S3-5. inputting the meteorological simulation test set M3 into the optimized first type of model and the second type of model to output a historical prediction sequence of the icing thickness of the power transmission line .

[0014] Preferably, the optimization algorithm in step S3-4 is a grid search or a genetic algorithm or a Bayesian optimization.

[0015] Further, step S4 includes the following steps: S4-1. the icing prediction mechanism model includes a Makkonen glaze icing growth model and a Jones rain icing growth model; S4-2. inputting the zonal wind u, the meridional wind v, the air temperature t, the air pressure p, the precipitation q in the meteorological simulation test set M3 and the power transmission line conductor diameter d0 in the power transmission line icing historical data set into the Makkonen glaze icing growth model to output a historical prediction sequence of glaze-type icing thickness HM; S4-3. inputting the zonal wind u, the meridional wind v, the air temperature t, the air pressure p, and the freezing rain z in the meteorological simulation test set M3 into the Jones rain icing growth model to output a historical prediction sequence of rain icing thickness HJ; S4-4. adding the historical prediction sequence of glaze-type icing thickness HM and the historical prediction sequence of rain icing thickness HJ to obtain a historical prediction sequence of the icing thickness of the power transmission line .

[0016] Further, step S5 includes the following steps: S5-1. the icing ensemble probability prediction machine learning model includes a Bayesian ridge regression or a random forest or a gradient boosting machine or a Gaussian process regression; S5-2. inputting the historical prediction sequence of the icing thickness of the power transmission line and the historical prediction sequence of the icing thickness of the power transmission line Merging the sequences , ; S5-3. Dividing the sequence into icing thickness prediction training set H1, icing thickness prediction validation set H2 in a time ratio of 8:2, and dividing the icing thickness test set R3 into icing thickness monitoring training set R1', icing thickness monitoring validation set R2' in a time ratio of 8:2; S5-4. Inputting the icing thickness prediction training set H1 as input features and the icing thickness monitoring training set R1' as training targets into the icing set probability prediction machine learning model to obtain the relationship model of the icing set probability prediction machine learning model; S5-5. Using the icing thickness prediction validation set H2 as input features and the icing thickness monitoring validation set R2' as validation targets to perform independent optimization on the hyperparameters of the relationship model of the icing set probability prediction machine learning model using an optimization algorithm, save the best parameter configuration of the relationship model of the icing set probability prediction machine learning model, and obtain the optimized icing set probability prediction machine learning model.

[0017] Preferably, the optimization algorithm in step S5-5 is Bayesian optimization or evolutionary algorithm or random search.

[0018] Further, step S6 includes the following steps: S6-1. Driving the small and micro scale wind field prediction model WRF-CALMET using real-time forecast field data of a future period, combining the icing monitoring data set , and the transmission line vector information shp in the icing monitoring data set to generate a meteorological prediction data set of the future period, which includes time s', transmission line name id', meridional wind u', zonal wind v', air temperature t', air pressure p', precipitation q', freezing rain z', and relative humidity r'; S6-2. Inputting the meteorological prediction data set of the future period into the optimized first type model and the second type model to output a future prediction sequence of transmission line icing thickness ; S6-3. Inputting the meteorological prediction data set of the future period into the icing prediction mechanism model to output a future prediction sequence of transmission line icing thickness ; S6-4. Comparing the future prediction sequence of transmission line icing thickness with the future prediction sequence of transmission line icing thickness The merged future prediction sequence of the ice thickness of the power transmission line , S6-5. The future prediction sequence of the ice thickness of the power transmission line is input into the optimized ice collection probability prediction machine learning model, and the ice thickness probability prediction result of the future period is output.

[0019] The beneficial effects of the present application are: (1) Through the small and micro scale wind field model, the prediction accuracy of the key meteorological field along the power transmission line under complex terrain is effectively improved, and the input quality of the ice prediction is improved from the source.

[0020] (2) The collection physical mechanism model and the data-driven artificial intelligence model are complementary, overcoming the principle limitations of single model, and significantly improving the adaptability and stability under different meteorological conditions.

[0021] (3) Through multi-model ensemble prediction, the prediction result with probability distribution is provided, the prediction uncertainty is quantified, the risk confidence level can be evaluated by the decision maker, and the leap from deterministic warning to probabilistic risk assessment is realized. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 The flow chart of the method of the present application. DETAILED DESCRIPTION

[0023] The present application will be further described below. Figure 1

[0024] A complex terrain power transmission line ice thickness probability prediction method based on multi-model ensemble includes: S1. Obtain the power transmission line ice thickness historical data set .

[0025] S2. Obtain the meteorological simulation data set same period as the power transmission line ice thickness historical data set .

[0026] S3. Establish an ice prediction machine learning model using the meteorological simulation data set and the power transmission line ice thickness historical data set , and obtain the power transmission line ice thickness historical prediction sequence according to the ice prediction machine learning model .

[0027] S4. Construct an ice prediction mechanism model according to the meteorological simulation data set , the power transmission line ice thickness historical data set ​​​A transmission line icing thickness historical prediction sequence is obtained by an icing prediction mechanism model .

[0028] S5. Utilize the transmission line icing thickness historical prediction sequence , the transmission line icing thickness historical prediction sequence and the transmission line icing historical data set An icing set probability prediction machine learning model is established, and the icing set probability prediction machine learning model is trained to obtain an optimized icing set probability prediction machine learning model.

[0029] S6. Construct a meteorological prediction data set of a future period , the meteorological prediction data set Input into the optimized icing prediction machine learning model, the icing prediction mechanism model and the optimized icing set probability prediction machine learning model, and output to obtain an icing thickness probability prediction result of the future period.

[0030] High-precision meteorological prediction data under complex terrain is provided through a small and micro scale wind field prediction mode; a plurality of icing thickness prediction models based on physical mechanisms and artificial intelligence are constructed based on the prediction data; the above-mentioned icing thickness prediction models are combined into an icing thickness probability prediction method, and the uncertainty of icing prediction is quantified. Through the method, the precision and robustness of icing thickness prediction under complex terrain conditions can be effectively improved, scientific basis with certainty and probability for power grid ice prevention and disaster reduction decision-making is provided, and the risk early warning and prevention and control ability of the power grid to respond to ice disasters is enhanced.

[0031] In an embodiment of the present application, 5 years of icing monitoring data are collected from a certain area in step S1 to obtain an icing monitoring data set The icing monitoring data include time s, transmission line name id, transmission line monitoring icing thickness h, transmission line conductor diameter d0 and transmission line vector information shp.

[0032] In an embodiment of the present application, step S2 includes the following steps: S2-1. Use the meteorological reanalysis data to drive the mesoscale weather mode WRF to generate a mesoscale meteorological field, and use the CALWRF pretreater to process the mesoscale meteorological field into meteorological data 3D.DAT readable by the microscale diagnostic mode CALMET.

[0033] S2-2. Collect the geophysical data required by the microscale diagnostic model CALMET, including SRTM DEM 90m resolution DEM terrain elevation data, ESRI Land Cover 10m resolution land use data, and sequentially using TERREL preprocessor, CTGPROC preprocessor and MAKEGEO preprocessor to process the geophysical data into geophysical data GEO.DAT readable by the microscale diagnostic model CALMET.

[0034] S2-3. For the target power transmission line area, the parameterization scheme in the microscale diagnostic model CALMET is set to turn off the terrain dynamics, turn on the O'Brien vertical velocity, turn on the downslope airflow effect, turn off the Froude number, the output time resolution is 1h, the output horizontal spatial resolution is 50m, and the vertical height is 10m, to obtain the small microscale wind field prediction model WRF-CALMET.

[0035] S2-4. Input the weather data 3D.DAT as the initial guess field and the geophysical data GEO.DAT as the environment field into the small microscale wind field prediction model WRF-CALMET, sequentially perform terrain dynamics, slope flow, terrain blocking effect, radiation minimization adjustment, interpolation smoothing, and vertical velocity calculation, to generate meteorological gridded simulation data, and use the same time s, the same power transmission line name id, and the same power transmission line vector information shp to extract the meridional wind u, the latitudinal wind v, the air temperature t, the air pressure p, the precipitation q, the freezing rain z, and the relative humidity r at the power transmission line from the meteorological gridded simulation data, to obtain the meteorological simulation data set including time s, power transmission line name id, meridional wind u, latitudinal wind v, air temperature t, air pressure p, precipitation q, freezing rain z, and relative humidity r.

[0036] Further preferably, the FNL global reanalysis data provided by the National Environmental Prediction Center of the United States is used as the initial field and side boundary condition driving data of the mesoscale weather model WRF. The numerical simulation adopts a double nested grid configuration, in which the horizontal resolution of the innermost grid is set to 9 kilometers. In the vertical direction, the number of model layers is set to 51 layers, and the pressure at the top of the model is 50 hPa. The microphysical process adopts the Thompson scheme, the long-wave and short-wave radiation transmission process adopts the RRTMG scheme, the cumulus convection parameterization adopts the Tiedtke scheme, the planetary boundary layer process adopts the YSU scheme, the near-surface layer process adopts the modified MM5 Monin-Obukhov scheme, and the land surface process adopts the Noah-MP land surface model.

[0037] In an embodiment of the present application, step S3 comprises the following steps: ​S3-1. The icing prediction machine learning model comprises a first type of model (a time series analysis principle type) and a second type of model (a nonlinear regression principle type), a number of a models are selected from the first type of model, and a number of b models are selected from the second type of model, a is greater than or equal to 1, and b is greater than or equal to 1, the first type of model comprises a long short-term memory network and a gated recurrent unit, and the second type of model comprises a support vector regression and a random forest. Preferably, the long short-term memory network is configured with two layers of hidden layers, and a dropout rate is set to 0.2, the support vector regression is set with a radial basis kernel function, a penalty factor C is set to 1.0, and a kernel function parameter gamma is set to scale.

[0038] S3-2. The meteorological simulation dataset is divided into a meteorological simulation training set M1, a meteorological simulation verification set M2, and a meteorological simulation test set M3 in a time ratio of 4:1:5, and the transmission line icing historical dataset is divided into a transmission line icing thickness training set R1, a transmission line icing thickness verification set R2, and a transmission line icing thickness test set R3 in a time ratio of 4:1:5.

[0039] S3-3. The meteorological simulation training set M1 is input as an input feature, and the transmission line icing thickness training set R1 is input as a training target into the first type of model and the second type of model respectively, and a relationship model of the first type of model and a relationship model of the second type of model are obtained respectively.

[0040] S3-4. An optimization algorithm is used to input the meteorological simulation verification set M2 as an input feature and the transmission line icing thickness verification set R2 as a verification target, the root mean square error on the meteorological simulation verification set M2 and the transmission line icing thickness verification set R2 is used as a fitness function, the hyperparameters of the relationship model of the first type of model are independently optimized through selection, crossover and mutation operations, the best parameter configuration of the relationship model of the first type of model is saved, and a trained first type of model is obtained, and the optimization algorithm is used to input the meteorological simulation verification set M2 as an input feature and the transmission line icing thickness verification set R2 as a verification target, the root mean square error on the meteorological simulation verification set M2 and the transmission line icing thickness verification set R2 is used as a fitness function, the hyperparameters of the relationship model of the second type of model are independently optimized through selection, crossover and mutation operations, the best parameter configuration of the relationship model of the second type of model is saved, and a trained second type of model is obtained.

[0041] S3-5. The meteorological simulation test set M3 is input into each of the optimized first type of model and the second type of model, and a transmission line icing thickness historical prediction sequence is output.

[0042] In this embodiment, preferably, the optimization algorithm in step S3-4 is a grid search or a genetic algorithm or a Bayesian optimization.

[0043] In an embodiment of the present application, step S4 comprises the following steps: S4-1. The icing prediction mechanism model comprises Makkonen glaze icing growth model, Jones rime icing growth model. The icing prediction mechanism model can provide icing thickness probability prediction results.

[0044] S4-2. The zonal wind u, the meridional wind v, the air temperature t, the air pressure p, the precipitation q and the power transmission line icing historical data set in the meteorological simulation test set M3 are input into the Makkonen glaze icing growth model, and a glaze type icing thickness historical prediction sequence HM is output.

[0045] S4-3. The zonal wind u, the meridional wind v, the air temperature t, the air pressure p, and the freezing rain z in the meteorological simulation test set M3 are input into the Jones rime icing growth model, and a rime type icing thickness historical prediction sequence HJ is output. S4-4. The glaze type icing thickness historical prediction sequence HM and the rime type icing thickness historical prediction sequence HJ are added to obtain a power transmission line icing thickness historical prediction sequence .

[0046] In an embodiment of the present application, step S5 comprises the following steps: S5-1. The icing set probability prediction machine learning model comprises Bayesian ridge regression or random forest or gradient boosting machine or Gaussian process regression. In this embodiment, random forest is preferably selected as the icing set probability prediction machine learning model, the number of decision trees of the random forest is set to 200, the number of features considered when splitting the node is set to the square root of the number of input features, the minimum number of samples required for node splitting is 5, and the minimum number of samples required for leaf node is 2.

[0047] S5-2. The power transmission line icing thickness historical prediction sequence is combined with the power transmission line icing thickness historical prediction sequence to obtain a sequence , .

[0048] S5-3. The sequence is divided into an icing thickness prediction training set H1 and an icing thickness prediction verification set H2 in a time ratio of 8:2, and the icing thickness test set R3 is divided into an icing thickness monitoring training set R1' and an icing thickness monitoring verification set R2' in a time ratio of 8:2.

[0049] S5-4. Input the icing thickness prediction training set H1 as the input feature and the icing thickness monitoring training set R1' as the training target into the icing set probability prediction machine learning model to obtain the relationship model of the icing set probability prediction machine learning model; S5-5. Use the icing thickness prediction verification set H2 as the input feature and the icing thickness monitoring verification set R2' as the verification target by using an optimization algorithm to obtain the continuous ranking probability score on the icing thickness prediction verification set H2 and the icing thickness monitoring verification set R2' as the optimization objective function, independently optimize the hyperparameters of the relationship model of the icing set probability prediction machine learning model, save the best parameter configuration of the relationship model of the icing set probability prediction machine learning model, and obtain the optimized icing set probability prediction machine learning model.

[0050] In this embodiment, preferably, the optimization algorithm in step S5-5 is Bayesian optimization or evolutionary algorithm or random search.

[0051] In one embodiment of the present application, step S6 includes the following steps: S6-1. Drive the small and micro scale wind field prediction model WRF-CALMET using the real-time forecast field data of the future period, combine the icing monitoring data set with the transmission line vector information shp in the icing monitoring data set to generate the meteorological prediction data set of the future period, and the meteorological prediction data set includes time s', transmission line name id', meridional wind u', zonal wind v', air temperature t', air pressure p', precipitation q', freezing rain z', and relative humidity r'.

[0052] S6-2. Input the meteorological prediction data set of the future period into the optimized first type model and the second type model to output the transmission line icing thickness future prediction sequence of the future period.

[0053] S6-3. Input the meteorological prediction data set of the future period into the icing prediction mechanism model to output the transmission line icing thickness future prediction sequence of the future period.

[0054] S6-4. Combine the transmission line icing thickness future prediction sequence with the transmission line icing thickness future prediction sequence to obtain the transmission line icing thickness future prediction sequence .

[0055] S6-5. Combine the transmission line icing thickness future prediction sequence The ice thickness probability prediction result of the future period is output by inputting into the optimized ice-covered set probability prediction machine learning model.

[0056] Finally, it should be noted that: the above only for the preferred embodiments of the present application, and not for limiting the present application, although the foregoing embodiments of the present application are described in detail, for those skilled in the art, it still can be modified, or the equivalent replacement of part of the technical features described in the foregoing embodiments. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included within the scope of the present application.

Claims

1. A method for probabilistic prediction of ice thickness on complex terrain power transmission lines based on a multi-modal set, characterized in that, Comprising: S1. Obtain a set of historical icing data for a power transmission line ; S2. Obtain a set of historical icing data for the power transmission line contemporaneous weather simulation data to obtain a set of weather simulation data ; S3. Utilizing meteorological simulation datasets with the transmission line icing history dataset establishing an icing prediction machine learning model, and obtaining a transmission line icing thickness history prediction sequence according to the icing prediction machine learning model ; S4. Constructing icing prediction mechanism model, according to meteorological simulation data set , transmission line icing historical data set and icing prediction mechanism model to obtain transmission line icing thickness historical prediction sequence ; S5. Utilizing a sequence of historical ice thickness predictions for a power transmission line , a sequence of historical ice thickness predictions for a power transmission line a historical dataset of ice cover for a power transmission line establishing an ice collection probability prediction machine learning model, training the ice collection probability prediction machine learning model to obtain an optimized ice collection probability prediction machine learning model; S6. Constructing a weather forecast dataset for a future period , weather forecast dataset Input into the optimized ice prediction machine learning model, the ice prediction mechanism model, the optimized ice set probability prediction machine learning model, and output to obtain the ice thickness probability forecast result for the future period.

2. The multi-modal ensemble based complex terrain transmission line ice thickness probabilistic forecasting method of claim 1, wherein: In step S1, 5 years of icing monitoring data from a certain area are collected to obtain an icing monitoring data set The icing monitoring data include time s, power line name id, power line monitored icing thickness h, power line conductor diameter d0, and power line vector information shp.

3. The multi-mode set based complex terrain transmission line ice thickness probabilistic forecasting method of claim 2, wherein, Step S2 comprises the following steps: S2-1. Generate a mesoscale meteorological field using a weather reanalysis dataset to drive a mesoscale weather model WRF, and use a CALWRF preprocessor to process the mesoscale meteorological field into meteorological data 3D.DAT readable by a microscale diagnostic model CALMET; S2-2. Collect geophysical data required by the microscale diagnostic model CALMET, the geophysical data including SRTM DEM 90m resolution DEM terrain elevation data, ESRI Land Cover 10m resolution land use data, and sequentially use a TERREL preprocessor, a CTGPROC preprocessor, and a MAKEGEO preprocessor to process the geophysical data into geophysical data GEO.DAT readable by the microscale diagnostic model CALMET; S2-3. For the target power transmission line area, set the parameterization scheme in the microscale diagnostic model CALMET to turn off terrain dynamics, turn on O'Brien vertical velocity, turn on downslope airflow effect, and turn off Froude number, output a time resolution of 1h, an output horizontal spatial resolution of 50m, and a vertical height of 10m, to obtain a small microscale wind field prediction model WRF-CALMET; S2-4. Input the meteorological data 3D.DAT as the initial guess field and the geophysical data GEO.DAT as the environmental field into the small-scale wind field prediction model WRF-CALMET, sequentially perform terrain dynamics, slope flow, terrain blocking effect, radiation minimization adjustment, interpolation smoothing, and vertical velocity calculation to generate meteorological gridded simulation data, and use the same as the meteorological simulation data set Extract the zonal wind u, meridional wind v, air temperature t, air pressure p, precipitation q, freezing rain z, and relative humidity r at the power transmission line from the meteorological gridded simulation data using the same time s, the same power transmission line name id, and the same power transmission line vector information shp. Include time s, power transmission line name id, zonal wind u, meridional wind v, air temperature t, air pressure p, precipitation q, freezing rain z, and relative humidity r.

4. The multi-mode set based complex terrain transmission line ice thickness probabilistic forecasting method of claim 2, wherein, Step S3 comprises the following steps: S3-1. The icing prediction machine learning model comprises a first type of model and a second type of model, a is selected from the first type of model, and b is selected from the second type of model, a is greater than or equal to 1, b is greater than or equal to 1, the first type of model comprises a long short-term memory network and a gated recurrent unit; the second type of model comprises a support vector regression and a random forest; S3-2. Meteorological simulation dataset The historical data set of power transmission line icing was divided into three sets: meteorological simulation training set M1, meteorological simulation validation set M2, and meteorological simulation test set M3, with a time ratio of 4:1:

5. The ice thickness h of the transmission line monitoring is divided into ice thickness training set R1, ice thickness validation set R2, and ice thickness test set R3 in a time ratio of 4:1:

5. S3-3. Input the meteorological simulation training set M1 as the input feature and the icing thickness training set R1 as the training target into the first type of model and the second type of model respectively, to obtain a relationship model of the first type of model and a relationship model of the second type of model respectively; S3-4. Use an optimization algorithm to use the meteorological simulation verification set M2 as the input feature and the icing thickness verification set R2 as the verification target to independently optimize the hyperparameters of the relationship model of the first type of model, save the best parameter configuration of the relationship model of the first type of model, obtain the trained first type of model, and use an optimization algorithm to use the meteorological simulation verification set M2 as the input feature and the icing thickness verification set R2 as the verification target to independently optimize the hyperparameters of the relationship model of the second type of model, save the best parameter configuration of the relationship model of the second type of model, and obtain the trained second type of model; S3-5. Input the weather simulation test set M3 into the optimized individual first-type model and second-type model, and output to obtain the historical prediction sequence of the ice thickness of the transmission line .

5. The multi-mode ensemble-based complex terrain power line ice accretion thickness probabilistic forecasting method of claim 4, wherein: The optimization algorithm in step S3-4 is a grid search or a genetic algorithm or a Bayesian optimization.

6. The multi-mode set based complex terrain power line ice thickness probabilistic forecasting method of claim 4, wherein, Step S4 comprises the following steps: S4-1. The icing prediction mechanism model comprises a Makkonen glaze icing growth model and a Jones rain icing growth model; S4-2. The zonal wind u, the meridional wind v, the air temperature t, the air pressure p, the precipitation q and the historical dataset of the transmission line icing in the weather simulation test set M3 are input into the Makkonen rime icing growth model, and the historical prediction sequence of the rime icing thickness HM is output. S4-3. Input the zonal wind u, the meridional wind v, the air temperature t, the air pressure p, and the frozen rain amount in the meteorological simulation test set M3 into the Jones rain icing growth model, and output to obtain a rain icing thickness historical prediction sequence HJ; S4-4. Add the rime-type ice accretion thickness historical prediction sequence HJ to the glaze-type ice accretion thickness historical prediction sequence HM to obtain a power transmission line ice accretion thickness historical prediction sequence H .

7. The multi-mode ensemble-based complex terrain power line ice accretion thickness probabilistic forecasting method of claim 4, wherein, Step S5 comprises the following steps: S5-1. The icing set probability prediction machine learning model comprises a Bayesian ridge regression or a random forest or a gradient boosting machine or a Gaussian process regression; S5-2. The sequence of historical ice accretion thickness predictions for the power transmission line with the sequence of historical ice accretion thickness predictions for the power transmission line resulting in the sequence , ; S5-3. The sequence The icing thickness test set R3 is divided into an icing thickness monitoring training set R1' and an icing thickness monitoring verification set R2' in a time ratio of 8:

2. S5-4. The icing set probability prediction machine learning model is inputted with the icing thickness prediction training set H1 as input features and the icing thickness monitoring training set R1' as training targets to obtain a relationship model of the icing set probability prediction machine learning model; S5-5. An optimization algorithm is used to input the icing thickness prediction verification set H2 as input features and the icing thickness monitoring verification set R2' as verification targets, to take the continuous ranking probability score on the icing thickness prediction verification set H2 and the icing thickness monitoring verification set R2' as an optimization objective function, to independently optimize the hyperparameters of the relationship model of the icing set probability prediction machine learning model, to save the best parameter configuration of the relationship model of the icing set probability prediction machine learning model, and to obtain an optimized icing set probability prediction machine learning model.

8. The multi-mode ensemble-based complex terrain power line ice accretion thickness probabilistic forecasting method of claim 7, wherein: The optimization algorithm in step S5-5 is a Bayesian optimization or an evolutionary algorithm or a random search.

9. The multi-mode ensemble-based complex terrain power line ice accretion thickness probabilistic forecasting method of claim 7, wherein, Step S6 comprises the following steps: S6-1. Using real-time forecast field data of future period to drive small and micro scale wind field prediction model WRF-CALMET, combined with icing monitoring dataset Vector information shp of power transmission line in the dataset Meteorological prediction dataset of future period Including time s', power transmission line name id', meridional wind u', zonal wind v', air temperature t', air pressure p', precipitation q', freezing rain z', relative humidity r' S6-2. The weather forecast dataset for future periods. The inputs are fed into the optimized Class I and Class II models, and the outputs are the future prediction sequences of transmission line icing thickness for future time periods. ; S6-3. The future time period's meteorological forecast dataset is input into the ice accretion prediction mechanism model, and the future prediction sequence of the power transmission line ice accretion thickness is output ; S6-4. Future prediction sequence of transmission line icing thickness Future prediction sequence of transmission line icing thickness Future prediction sequence of transmission line icing thickness , ; S6-5. Future prediction sequence of transmission line icing thickness The input is put into the optimized icing set probability prediction machine learning model, and the output is obtained.

Images