Intelligent glaze forming weather identification method based on multi-source meteorological data fusion

Abstract

The invention belongs to the technical field of intelligent weather identification, and discloses an intelligent glaze forming weather identification method based on multi-source meteorological data fusion. The method comprises the steps of S3, performing data preprocessing on a glaze forming data set and a multi-source depth data set to obtain a feature vector for glaze forming weather recognition, S4, analyzing the feature vector based on a glaze recognition model to obtain a glaze forming value, and S5, judging the glaze influence degree based on the glaze forming value, the method comprises the following steps: S1, carrying out on-site manual observation on the basis of a glaze forming value, and generating an intelligent response report according to the glaze influence degree, S6, carrying out quantitative interpretation on the basis of the glaze forming value to obtain an auxiliary recognition report, S7, obtaining a real-time reporting result on the basis of on-site manual observation, and S8, forming an active learning mechanism according to the real-time reporting result and the glaze forming value. The method has the remarkable advantages that the glaze risk degree calculation accuracy degree is high, the auxiliary effect is good, and the system self-optimization capacity is high.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of weather intelligent identification, more particularly, the present application relates to a rain-ice formation weather intelligent identification method based on multi-source meteorological data fusion. BACKGROUND

[0002] Rain-ice, commonly known as "tree hanging", also called ice, is a kind of glass or transparent hard ice layer formed by freezing of supercooled liquid precipitation (freezing rain) after colliding with ground objects. Its outer surface is smooth or slightly protruding. The rain that forms rain-ice is called freezing rain. Due to the smoothness of rain-ice itself and the adhesion to the outer surface of exposed objects, it is easy to cause harm to roads or power lines.

[0003] Traditional rain-ice formation weather intelligent identification systems mostly rely on single data for identification, and the setting of sensors is easily affected by extreme weather, resulting in poor data transmission or data collection errors. Therefore, the identification data accuracy is poor. Therefore, how to effectively combine multi-source meteorological data to accurately calculate the formation of rain-ice and reduce the disaster impact of rain-ice has become a major natural problem facing the current society.

[0004] In view of this, the present application provides a rain-ice formation weather intelligent identification method based on multi-source meteorological data fusion to solve the above problems. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, in order to achieve the above-mentioned purpose, the present application provides the following technical scheme, the method comprises: S1: collecting rain-ice formation data set based on sensor network; this step specifically includes collecting rain-ice formation data set through sensor network, and the rain-ice formation data set includes environmental temperature data, environmental humidity data, environmental wind speed data and precipitation type data; S2: collecting multi-source depth data set based on sensor network; this step specifically includes collecting multi-source depth data set through sensor network, and the multi-source depth data set includes cloud layer temperature data, water vapor flux data, terrain height data and heat island intensity data; S3: data preprocessing of rain-ice formation data set and multi-source depth data set to obtain feature vector for rain-ice formation weather identification; this step specifically includes data cleaning, data normalization, data feature extraction and data interaction extraction of rain-ice formation data set and multi-source depth data set, which provides more convenient data support for subsequent rain-ice formation value calculation, effectively reduces the data processing pressure of step S4 and improves the accuracy of data calculation; S4: Analyzing the feature vector based on the glaze identification model to obtain a glaze formation value; this step specifically includes calculating the feature vector by using the glaze identification model, so that the system can effectively distinguish from the traditional static model calculation method, thereby fully exploiting the interaction characteristics between multiple data sources, greatly reducing the glaze misjudgment caused by single data source error, at the same time, with the explicit expression of the physical mechanism between multiple data sources, it can also effectively solve the problems of low recognition accuracy and weak data reference brought by traditional statistical methods or pure data driven methods, thereby further enhancing the accuracy of the system in identifying glaze formation weather; S5: Judging the glaze influence degree based on the glaze formation value, and generating an intelligent response report according to the glaze influence degree; this step specifically includes classifying the glaze formation value based on the influence threshold interval, and generating the corresponding response measures according to the judgment result, so that the system can effectively break through the barrier between data and the physical world, realize the conversion of data into practical countermeasures in the physical world, thereby further improving the versatility and practicality of the system, greatly reducing the low utilization rate of the system caused by the single work of traditional identification system and insufficient data mining; S6: Quantitative interpretation based on the glaze formation value to obtain an auxiliary identification report; this step specifically includes reverse deduction of the contribution of multiple data sources to the glaze formation value based on the glaze formation value, so that the system can effectively handle the "black box" problem of traditional identification system, provide data support for auxiliary staff to intuitively seek the main contribution data of glaze risk degree, and further make the system have decision basis support while providing effective and accurate data basis for process traceability judgment, thereby greatly enhancing the transparency and credibility of the system; S7: Obtaining real-time reporting results based on field manual observation; this step specifically includes providing real data feedback to the system through manual inspection observation, so that the system can dynamically adjust the model threshold and trigger data re-labeling process according to the real-time reporting results, and further optimize the calculation accuracy of the system; S8: Forming an active learning mechanism according to the real-time reporting results and the glaze formation value; this step specifically includes correcting the system parameters in real time based on the real-time reporting results, thereby forming an active learning mechanism for the system, effectively solving the problems of insufficient data coverage and distribution deviation of traditional identification system, and providing physical world assistance for the accuracy of the system in calculating glaze risk degree, S9: Storing the system data set to the database, displaying the auxiliary identification report through the visualization panel, processing the intelligent response report, and outputting according to the processing result; Further, the step S1 includes: Collecting the environmental temperature value in the specified area by using the digital temperature sensor to obtain the environmental temperature data; The environmental humidity data is obtained by collecting the environmental humidity value in the specified area through the capacitive humidity sensor. The environmental wind speed data is obtained by collecting the wind speed value in the specified area through the ultrasonic anemometer. The precipitation type data is obtained by collecting the precipitation type in the specified area through the laser precipitation phenomenon instrument, and assigning values to the precipitation type. Further, the step S2 comprises: The cloud layer temperature data is obtained by collecting the cloud layer bottom temperature value in the specified area through the infrared ceilometer. The water vapor flux data is obtained by collecting the atmospheric column water vapor content in the specified area through the microwave radiometer, and combining the horizontal wind speed and wind direction data to calculate the flux value. The terrain height data is obtained by collecting the terrain height value in the specified area through the RTK-GPS receiver. The heat island intensity data is obtained by collecting the surface temperature in the specified area through the thermal infrared imager. Further, the step S3 comprises: S3.1: Data cleaning is performed on all sub-data items in the basic data set by missing value filling, and all sub-data items in the basic data set are normalized to the range of [0, 1] based on the normalization formula. S3.2: Identify the environmental temperature data and precipitation type data. When the environmental temperature data is less than 0 degrees Celsius and the precipitation type data is greater than or equal to 2, the cold rain feature data with a value of 1 is obtained , otherwise, the cold rain feature data with a value of 0 is obtained. S3.3: Calculate the wind-cold feature data based on the environmental temperature data and the environmental wind speed data. The specific formula is: ; The wind-cold feature data is obtained , wherein is the environmental temperature data, is the environmental wind speed data. S3.4: Calculate the water vapor feature based on the environmental wind speed data and the water vapor flux data. The specific formula is: , the water vapor feature data is obtained , wherein is the divergence operator, is the water vapor flux data. S3.5: Calculate the terrain lifting feature based on the terrain height data. The specific formula is: ; The terrain lifting feature data is obtained , wherein To maximize the function, For terrain height data, This is the baseline terrain height data; S3.6: By calculating ambient temperature data and 0.5 times the heat island intensity data The sum of these values ​​yields the corrected feature data. ; S3.7: Subtract the ambient temperature data from the cloud temperature data to obtain the temperature difference characteristic data. ; S3.8: Package cold rain feature data, wind chill feature data, water vapor feature data, topographic uplift feature data, correction feature data, and temperature difference feature data to obtain feature vectors; Further, step S4 includes: S4.1: Based on the historical feature dataset, the nonlinear hybrid mechanism model is trained. The specific formula for the nonlinear hybrid mechanism model is as follows: ; Obtain the first rime formation value ,in, For characteristic collaborative gating factors, It is a thermodynamic oscillation modulator. The phase transition energy threshold, For the number of historical windows, For the first Each historical weighting factor For the first A topographical historical memory core Activation function, As a dynamic weighting factor, For partial derivatives, For the first Sub-feature data in a feature vector For timestamps, It is the hyperbolic tangent function. For cloud cold energy integrator; Step S4.2: Decompose the computational formulas for the feature co-gating factor, thermodynamic oscillation modulator, phase change energy threshold, terrain history memory kernel, and cloud cold energy integrator in step S4.1. The specific expression formula set is as follows: ; in, For the Sigmoid function, For the first The mean of the eigenvectors of the terms. For the first The standard deviation of the eigenvectors of the term, For the first A time decay factor, is maximum terrain height data, is minimum terrain height data, is the first history of rime formation value, is the number of spatial grids; Step S4.3: When the number of times of training the nonlinear mixing mechanism model in step S4.1 reaches the preset iteration number, output the rime identification model obtained; Step S4.4: Input the feature vector into the rime identification model to output the rime formation value obtained; Step S4.5: Output the rime formation value to step S5 and step S6; Further, step S5 comprises: Based on the influence threshold interval (W1, W2), the rime formation value is substituted into the influence threshold interval for comparison; When the rime formation value is less than W1, a no-risk report is generated, when the rime formation value is greater than or equal to W1 and less than W2, a medium-risk report is generated, and when the rime formation value is greater than or equal to W2, a heavy-risk report is generated; The no-risk report includes that the current rime danger level is low, please work according to the preset work procedure, and ensure continuous monitoring of the weather state; The medium-risk report includes that the current rime danger level is medium, please start the traffic warning measure, and start the unmanned aerial vehicle to inspect the important power line section; The heavy-risk report includes that the current rime danger level is high, please publish the traffic ban warning, close the traffic permission of the bridge section, and start the direct current ice melting device of the power line; Pack the no-risk report, the medium-risk report and the heavy-risk report to obtain an intelligent response report; Further, the step S6 comprises: Based on the rime formation value and the feature vector, the contribution rate of the key feature data is calculated, and the specific formula is: ; the first feature data contribution rate , wherein, is the weight factor of the first feature data, is the feature vector of the first feature data, is the rime formation value; Pack all the contribution rates, combine them from large to small according to the contribution rate values, and combine the corresponding feature data symbols to obtain an auxiliary identification report; Further, the step S7 comprises: Based on setting artificial inspection observation mechanism, the inspector uploads actual rime thickness data and influence range score data through mobile terminal APP; Packing actual rime thickness data and influence range score data, real-time reporting results are obtained; Further, the step S8 comprises: Based on rime formation value and real-time reporting results; When the absolute value of rime formation value minus influence range score data is greater than 2, a weight factor adjustment report is generated; When the actual rime thickness data is greater than or equal to 5, a characteristic weight factor adjustment report is generated; The weight factor adjustment report comprises a set of instructions for increasing or decreasing 5% weight factor ; The characteristic weight factor adjustment report comprises a set of instructions for increasing 3% cold rain characteristic data weight factor; The weight factor adjustment report is transmitted to step S6, and the characteristic weight factor adjustment report is sent to step S4.1; Further, the step S9 comprises: The system data set comprises rime formation data set, multi-source depth data set, feature vector, rime formation value, intelligent response report, auxiliary identification report, real-time reporting result and parameter adjustment report; When the intelligent response report is a no-risk report, the no-risk report is sent to the staff mailbox receiving end through the mail mode; When the intelligent response report is a medium-risk report, the medium-risk report is sent to the staff mailbox receiving end through the mail mode, and the staff is reminded to pay attention to check the mail through the short message mode; When the intelligent response report is a heavy-risk report, the heavy-risk report is sent to the staff mailbox receiving end through the mail mode, and the staff is reminded to pay attention to check the mail through the telephone mode.

[0006] The technical effects and advantages of the rime formation weather intelligent identification method based on multi-source meteorological data fusion of the application are: The application collects a rime formation data set based on a sensor network, collects a multi-source depth data set based on a sensor network, performs data preprocessing on the rime formation data set and the multi-source depth data set, obtains a feature vector for rime formation weather identification, analyzes the feature vector based on a rime identification model, obtains a rime formation value, judges the rime influence degree based on the rime formation value, generates an intelligent response report according to the rime influence degree, performs quantitative interpretation based on the rime formation value to obtain an auxiliary identification report, obtains real-time reporting results based on on-site manual observation, forms an active learning mechanism according to the real-time reporting results and the rime formation value, stores the system data set to a database, displays the auxiliary identification report through a visual panel, processes the intelligent response report, and outputs according to the processing result, so that the system can accurately calculate the danger degree of rime under the dynamic fluctuation of multiple data, provide a data basis for the timely response of staff, in addition, the intelligent response report obtained by the grading judgment of the rime formation value can effectively assist the staff to implement accurate and effective response measures according to the different rime risk degrees, greatly reducing the time difference from data identification to decision implementation, providing valuable time cost savings for disaster response, at the same time, through the auxiliary identification report, the system can have the ability of data tracing, and show the staff the leading data of rime formation, so as to further achieve the purpose of enabling the staff to respond to the measures according to the leading data, and through the real-time reporting results, the system can have the ability of active learning, realize the effect of active optimization of calculation accuracy, greatly reduce the probability of large output data error of the system caused by objective factors such as insufficient sensor coverage, in general, the application has the significant advantages of high accuracy of rime risk degree calculation, good auxiliary effect and strong self-optimization ability of the system. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 The application is a schematic diagram of a rime formation weather intelligent identification method based on multi-source meteorological data fusion. DETAILED DESCRIPTION

[0008] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0009] Embodiment 1 Please refer to Figure 1 The rime formation weather intelligent identification method based on multi-source meteorological data fusion described in this embodiment comprises: S1: Collecting a glaze formation dataset based on a sensor network; this step specifically includes collecting a glaze formation dataset through a sensor network, and the glaze formation dataset includes environmental temperature data, environmental humidity data, environmental wind speed data, and precipitation type data; S2: Collecting a multi-source depth dataset based on a sensor network; this step specifically includes collecting a multi-source depth dataset through a sensor network, and the multi-source depth dataset includes cloud layer temperature data, water vapor flux data, terrain height data, and heat island intensity data; S3: Data preprocessing of the glaze formation dataset and the multi-source depth dataset to obtain a feature vector for glaze formation weather recognition; this step specifically includes data cleaning, data normalization, data feature extraction, and data interaction extraction of the glaze formation dataset and the multi-source depth dataset, providing more convenient data support for subsequent glaze formation value calculation, effectively reducing the data processing pressure of step S4 and improving the accuracy of data calculation; S4: Analyzing the feature vector based on a glaze recognition model to obtain a glaze formation value; this step specifically includes calculating the feature vector using a glaze recognition model, enabling the system to effectively distinguish from traditional static model calculation methods, thereby fully exploiting the interaction characteristics between multi-source data, greatly reducing the glaze misjudgment caused by single data source error, and at the same time, with the explicit expression of the physical mechanism between multi-source data, it can also effectively solve the problems of low recognition accuracy and weak data reference brought by traditional statistical methods or pure data-driven methods, thereby further enhancing the accuracy of the system in recognizing glaze formation weather; S5: Judging the glaze influence degree based on the glaze formation value and generating an intelligent response report according to the glaze influence degree; this step specifically includes classifying the glaze formation value based on the influence threshold interval, and generating corresponding response measures according to the judgment result, so that the system can effectively break through the barrier between data and the physical world, realize the conversion of data into practical countermeasures in the physical world, thereby further improving the versatility and practicality of the system, greatly reducing the low utilization rate of the system caused by the single work of traditional recognition system and insufficient data mining; S6: Quantitative interpretation based on the glaze formation value to obtain an auxiliary recognition report; this step specifically includes reverse deduction of the contribution of multi-source data to the glaze formation value based on the glaze formation value, so that the system can effectively handle the "black box" problem of traditional recognition systems, providing data support for auxiliary staff to intuitively seek the main contribution data of glaze danger degree, and further enabling the system to provide effective and accurate data basis for process traceability judgment while supporting decision-making, thereby greatly enhancing the transparency and credibility of the system; S7: Based on the on-site manual observation, the real-time reporting result is obtained; this step specifically includes providing real data feedback for the system through manual inspection observation, so that the system can dynamically adjust the model threshold and trigger data re-labeling process according to the real-time reporting result, and further optimize the calculation accuracy of the system; S8: Forming an active learning mechanism according to the real-time reporting result and the rime formation value; this step specifically includes correcting the system parameters in real time based on the real-time reporting result, thereby forming an active learning mechanism for the system, effectively solving the problems of insufficient data coverage and distribution deviation of the traditional recognition system, and providing a physical world auxiliary role for the accuracy of the system in calculating the rime risk degree, S9: The system data set is stored in the database, the auxiliary identification report is displayed through the visual panel, the intelligent response report is processed, and the output is carried out according to the processing result; The core of the application is that the rime formation data set and the multi-source depth data set are obtained through the sensor network, and data cleaning, data normalization and feature extraction are carried out, and the calculation of the rime formation value is realized based on the extracted feature vector. Specifically, the basic data support for feature extraction of step S3 is realized through the basic data acquisition provided by steps S1 and S2, step S4 trains the non-linear mixing mechanism model through the feature vector of step S3, and obtains the rime recognition model after training. By inputting the feature vector, the rime formation value reflecting the rime danger degree is obtained, step S5 provides corresponding countermeasures for the staff based on the different degrees of rime formation value, and the rime formation value provides core data support in the auxiliary identification feature data contribution rate of step S6 and the active learning mechanism of step S8. The real-time reporting result collected in step S7 provides real data support for system error correction, and finally the data obtained in all steps is stored or transmitted through step S9, so that the system can finally complete the closed loop chain of data acquisition→data processing→data calculation→calculation result application→system parameter error correction→data storage and output. Further, the step S1 comprises: The environment temperature data is obtained by collecting the environment temperature value in the specified area through the digital temperature sensor; The environment humidity data is obtained by collecting the environment humidity value in the specified area through the capacitive humidity sensor; The environment wind speed data is obtained by collecting the wind speed value in the specified area through the ultrasonic anemometer; The precipitation type data is obtained by collecting the precipitation type in the specified area through the laser precipitation phenomenon instrument, and the precipitation type is valued; It needs to be explained that the precipitation type includes but is not limited to no precipitation, rain, freezing rain or snow, for example, when the precipitation type is no precipitation, the precipitation type data is assigned a value of 0, when the precipitation type is rain, the precipitation type data is assigned a value of 1, and the assignment of the precipitation type data is based on the precipitation type and so on; The core of this embodiment is that the sensor network collects a variety of glaze formation directly related data, which provides intuitive data support for the feature extraction of subsequent step S3 and the glaze formation value calculation of step S4; Further, the step S2 comprises: Through the infrared cloud height instrument, the cloud layer bottom temperature value in the specified area is collected to obtain the cloud layer temperature data; Through the microwave radiometer, the atmospheric column water vapor content in the specified area is collected, and the flux value is calculated combined with the horizontal wind speed and wind direction data to obtain the water vapor flux data; Through the RTK-GPS receiver, the terrain height value in the specified area is collected to obtain the terrain height data; Through the thermal infrared imager, the ground surface temperature in the specified area is collected to obtain the heat island intensity data; The core of this embodiment is that the sensor network collects glaze formation related in-depth data, which provides auxiliary in-depth effect for the glaze formation data set of step S1, and provides multi-source and in-depth data support for the feature extraction, feature interaction of step S3 and the glaze formation value calculation of step S4; Further, the step S3 comprises: S3.1: Through missing value filling, the data cleaning is performed on all sub-data items in the basic data set, and all sub-data items in the basic data set are normalized to the range of [0, 1] based on the normalization formula; It needs to be explained that the basic data set includes the glaze formation data set and the multi-source in-depth data set; the missing value filling means that when the data of a certain sub-data item is missing, the missing data is supplemented using the mean value of the data values of the previous and next adjacent points of the missing data; the specific expression formula of the normalization formula is: , to obtain the normalized value , wherein is any sub-data item of the basic data, is the historical maximum value of the arbitrary sub-data item, is the historical minimum value of the arbitrary sub-data item; the normalization is used to eliminate the dimensions of all sub-data items in the basic data set; S3.2: Identify the ambient temperature data and the precipitation type data, when the ambient temperature data is less than 0 degrees Celsius and the precipitation type data is greater than or equal to 2, obtain the cold rain feature data with a value of 1 , otherwise, the cold rain feature data with a value of 0 is obtained; It needs to be explained that the environmental temperature data and the precipitation type data in step S3.2 are original data values without data normalization; S3.3: Calculate the wind-cold feature data based on the environmental temperature data and the environmental wind speed data, and the specific formula is: ; Get the wind-cold feature data , wherein, is the environmental temperature data, is the environmental wind speed data; S3.4: Calculate the water vapor feature based on the environmental wind speed data and the water vapor flux data, and the specific formula is: , get the water vapor feature data , wherein, is the divergence operator, is the water vapor flux data; It needs to be explained that the divergence operator is used to calculate the divergence of the vector field; S3.5: Calculate the terrain lifting feature based on the terrain height data, and the specific formula is: ; Get the terrain lifting feature data , wherein, is the maximum function, is the terrain height data, is the baseline terrain height data; It needs to be explained that the maximum function is used to select the maximum value of the two values in the adjacent brackets as the output value; S3.6: Calculate the sum of the environmental temperature data and 0.5 times the heat island intensity data , get the modified feature data ; S3.7: Subtract the cloud temperature data from the environmental temperature data to get the temperature difference feature data ; It needs to be explained that all the sub-data items in the basic data set involved in steps S3.3 to S3.7 are data values after data cleaning and data normalization; S3.8: Pack the cold rain feature data, the wind-cold feature data, the water vapor feature data, the terrain lifting feature data, the modified feature data and the temperature difference feature data to get the feature vector; The core of the embodiment is that, by data cleaning and data normalization on the glaze formation data set and the multi-source depth data set, the data can be more complete and the dimension can be eliminated to lay the foundation for subsequent feature extraction and interactive feature extraction, and through the feature vector obtained by the extraction of the multi-source data and the interactive feature extraction, the system can perform pre-processing on the data before calculating the glaze formation value, reduce the data processing time for subsequent system calculation, so that the system can not only ensure the accuracy of data calculation, but also effectively reduce the required time length of data calculation, thereby reducing the time cost of data processing; Further, the step S4 comprises: S4.1: based on the historical feature data set, a nonlinear mixed mechanism model is trained, and the specific expression formula of the nonlinear mixed mechanism model is: ; to obtain the first glaze formation value , wherein, is a feature cooperative gate factor, is a thermal oscillation modulator, is a phase change energy threshold, is a historical window number, is the th historical weight factor, is the th terrain history memory kernel activation function, is a dynamic weight factor, is a partial derivative, is a sub-feature data in the th feature vector, is a timestamp, is a hyperbolic tangent function, is a cloud layer cold energy integrator; It should be explained that the hyperbolic tangent function is used to constrain the output value in the immediately adjacent brackets within the range of [-1, 1]; Step S4.2: decompose the feature cooperative gate factor, the thermal oscillation modulator, the phase change energy threshold, the terrain history memory kernel and the cloud layer cold energy integrator in step S4.1 into calculation formulas, and the specific expression formula group is: ; wherein, is a Sigmoid function, is a mean of the th feature vector, is a standard deviation of the th feature vector, is the th time decay factor, is maximum terrain height data, is minimum terrain height data, is the first is the historical rime formation value, is the number of spatial grids; It needs to be explained that the Sigmoid function is used to constrain the output value in the immediately adjacent parentheses within the range of (0, 1); Step S4.3: When the number of times of training the nonlinear mixing mechanism model in step S4.1 reaches the preset iteration number, output the rime identification model; Step S4.4: Input the feature vector into the rime identification model to output the rime formation value; Step S4.5: Output the rime formation value to step S5 and step S6; The core of this embodiment is that the nonlinear mixing mechanism model is trained by the historical feature vector, and the rime identification model for calculating the feature vector is output, and finally the rime formation value reflecting the rime risk level is output. Specifically, the nonlinear mixing mechanism model is trained by the historical feature data set, and the rime identification model is obtained after training, and then the feature vector is input to obtain the rime formation value, so that the system can accurately and effectively quantify the rime risk level. Since the training is combined with the feature data of multiple data sources, it is ensured that the contribution of each feature data and the basic data conforms to the actual physical law, thereby effectively avoiding the data trust crisis caused by traditional single data source, so that the system can realize the quantitative mapping from multi-source environmental data to rime risk, and further balance the relationship between physical mechanism and data-driven, and further improve the credibility and accuracy of the rime formation value; Further, step S5 includes: Based on the influence threshold interval (W1, W2), the rime formation value is substituted into the influence threshold interval for comparison; It needs to be explained that the influence threshold interval is obtained by manual decision and input into the system; When the rime formation value is less than W1, a no-risk report is generated, when the rime formation value is greater than or equal to W1 and less than W2, a medium-risk report is generated, and when the rime formation value is greater than or equal to W2, a heavy-risk report is generated; The no-risk report includes that the current rime danger level is low, please work according to the preset work procedure, and ensure continuous monitoring of the weather state; The medium-risk report includes that the current rime danger level is medium, please start the traffic warning measure, and start the unmanned aerial vehicle to inspect the important power line section; The heavy-risk report includes that the current rime danger level is high, please publish the traffic ban warning, close the traffic permission of the bridge section, and start the direct current ice melting device of the power line. packaging no-risk reports, medium-risk reports and high-risk reports to obtain intelligent response reports; The core of this embodiment is that the icing formation value is processed by the artificial determined influence threshold interval to obtain corresponding countermeasures. Specifically, the theoretical data value calculated by the system is converted into direct and intuitive action instructions based on the influence threshold interval, and a strict risk control system is established, so that the system can greatly reduce the over-response of the low-risk area according to the different degrees of icing formation value, thereby greatly reducing the countermeasure cost of icing risk conditions. At the same time, through the risk control system, personnel and property safety can be guaranteed to the greatest extent, and cross-departmental collaborative response can be realized, thereby further improving the multi-use of system data; Further, the step S6 comprises: Based on the icing formation value and the feature vector, the contribution rate of the key feature data is calculated, and the specific formula is: ; obtain the contribution rate of the first feature data , wherein, is the weight factor of the first feature data, is the feature vector of the first feature data, is the icing formation value; Pack all the contribution rates, combine them in descending order of the contribution rate values, and combine the corresponding feature data symbols to obtain an auxiliary identification report; The core of this embodiment is that the contribution rate of a single feature data to the icing formation value is calculated by the calculation formula, and the auxiliary identification report is obtained by integration. Specifically, the contribution rate of a single feature data to the icing formation value is calculated based on the icing formation value and the feature vector, and then all the contribution rates are integrated, so that the system can provide intuitive dominant factor display for the staff according to the size of the contribution rate, thereby further providing data support for the staff to implement defense countermeasures, and effectively solving the data traceability problem in the traditional identification system; Further, the step S7 comprises: Based on the setting of the artificial patrol observation mechanism, the patrol personnel uploads the actual icing thickness data and the influence range score data through the mobile terminal APP; It should be explained that the unit of the icing thickness data is millimeter; and the influence range score data is the percentage of the icing coverage area to the total area of the specified area determined by artificial judgment; Pack the actual icing thickness data and the influence range score data to obtain real-time reporting results; The core of this embodiment is that by establishing an artificial inspection observation mechanism, data is obtained through on-site inspection by an inspector, so that the system can establish a ground true value verification channel and obtain system blind area information, avoid data loss of local icing of the bridge caused by difficulty of coverage of the sensor, and further improve the accuracy of the system in calculating the danger of rime by supplementing the data through system calculation and artificial inspection, thereby ensuring high coincidence between the data and the actual state; Further, the step S8 comprises: based on the rime formation value and the real-time reporting result; When the absolute value of the rime formation value minus the influence range score data is greater than 2, a weight factor adjustment report is generated; When the actual rime thickness data is greater than or equal to 5, a feature weight factor adjustment report is generated; The weight factor adjustment report comprises a set of instructions for increasing or decreasing the weight factor by 5%; The feature weight factor adjustment report comprises a set of instructions for increasing the weight factor of the cold rain feature data by 3%; The feature weight factor adjustment report comprises a set of instructions for increasing the weight factor of the cold rain feature data by 3%; The weight factor adjustment report is transmitted to step S6, and the feature weight factor adjustment report is sent to step S4.1; The core of this embodiment is that by interacting the rime formation value and the real-time reporting result, and obtaining the weight factor adjustment report and the feature weight factor adjustment report according to the judgment result, specifically, through system calculation and artificial observation, the system can realize self-evolution, dynamically adjust the specific parameters of the system calculation, and provide more accurate framework support for the calculation of the next period; Further, the step S9 comprises: The system data set comprises a rime formation data set, a multi-source depth data set, a feature vector, a rime formation value, an intelligent response report, an auxiliary identification report, a real-time reporting result, and a parameter adjustment report; It should be explained that the parameter adjustment report comprises the weight factor adjustment report and the feature weight factor adjustment report; When the intelligent response report is a no-risk report, the no-risk report is sent to the staff mailbox receiving end through the email mode; When the intelligent response report is a medium-risk report, the medium-risk report is sent to the staff mailbox receiving end through the email mode, and the staff is reminded to pay attention to check the email through the short message mode; When the intelligent response report is a high-risk report, the high-risk report is sent to the staff mailbox receiving end through the email mode, and the staff is reminded to pay attention to check the email through the telephone mode; The beneficial effects of the embodiment are that the raindrop formation data set is collected based on the sensor network, the multi-source depth data set is collected based on the sensor network, the raindrop formation data set and the multi-source depth data set are preprocessed to obtain a feature vector for raindrop formation weather identification, the feature vector is analyzed based on a raindrop identification model to obtain a raindrop formation value, the raindrop influence degree is judged based on the raindrop formation value, the intelligent response report is generated according to the raindrop influence degree, the quantitative explanation is performed based on the raindrop formation value to obtain an auxiliary identification report, the real-time reporting result is obtained based on the field manual observation, the active learning mechanism is formed according to the real-time reporting result and the raindrop formation value, the system data set is stored in the database, the auxiliary identification report is displayed through the visual panel, the intelligent response report is processed, and the output is performed according to the processing result, so that the system can accurately calculate the danger degree of the raindrop under the dynamic fluctuation of multiple data, provide a data basis for the timely response of the staff, in addition, the intelligent response report obtained through the grading judgment of the raindrop formation value can effectively assist the staff to quickly implement accurate and effective response measures according to the different raindrop risk degrees, greatly reduce the time difference from data identification to decision implementation, provide valuable time cost savings for disaster response, at the same time, through the auxiliary identification report, the system can have the data tracing capability, and the leading data of the raindrop formation is displayed to the staff, so that the staff can respond to the measures according to the leading data, and the system can have the active learning capability, realize the effect of active optimization of the calculation accuracy, greatly reduce the probability of large output data error of the system caused by objective factors such as insufficient sensor coverage, and overall, the application has the significant advantages of high raindrop risk degree calculation accuracy, good auxiliary effect and strong system self-optimization capability.

[0010] It is obvious to a person skilled in the art that the application is not limited to the details of the above exemplary embodiments, but can be implemented in other concrete forms without departing from the spirit or essential characteristics of the application.

[0011] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, the scope of the application is defined by the appended claims rather than the above description, and therefore all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the application. Any additional reference signs in the claims should not be considered as limiting the claims to which they relate.

[0012] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The plurality of units or devices stated in the system claims can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not mean any particular order.

[0013] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the present application but not to limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced without departing from the spirit and scope of the present application.

Claims

1. A method for intelligent identification of rime formation weather based on multi-source meteorological data fusion, characterized in that, The method includes: S1: Collect data on rime ice formation based on sensor networks; S2: Based on sensor networks, collect multi-source depth datasets; S3: Perform data preprocessing on the rime formation dataset and multi-source depth dataset to obtain feature vectors for rime formation weather identification; S4: Based on the rime identification model, the feature vector is analyzed to obtain the rime formation value; S5: Determine the degree of impact of rime ice based on the rime ice formation value, and generate an intelligent response report based on the degree of impact of rime ice; S6: Quantitatively interpret the rime formation value to obtain an auxiliary identification report; S7: Real-time reporting results are obtained based on on-site manual observation; S8: Based on real-time reporting results and rime formation values, an active learning mechanism is formed; S9: Store the system dataset in the database, display the assisted recognition report through the visualization panel, process the intelligent response report, and output the results based on the processing.

2. The intelligent identification method for rime formation weather based on multi-source meteorological data fusion according to claim 1, characterized in that, S1 includes: Ambient temperature data is obtained by collecting ambient temperature values ​​within a designated area using a digital temperature sensor. A capacitive humidity sensor is used to collect ambient humidity values ​​within a specified area to obtain ambient humidity data. By using an ultrasonic anemometer, wind speed values ​​are collected within a designated area to obtain environmental wind speed data. The precipitation type data is obtained by collecting precipitation type data within a designated area using a laser precipitation instrument and assigning values ​​to the precipitation types.

3. The intelligent identification method for rime formation weather based on multi-source meteorological data fusion according to claim 1, characterized in that, S2 includes: By using an infrared ceilometer, the temperature values ​​at the bottom of the clouds in a designated area are collected to obtain cloud temperature data. The atmospheric water vapor content in a designated area is collected using a microwave radiometer, and the flux value is calculated by combining the horizontal wind speed and wind direction data to obtain water vapor flux data. The terrain elevation data is obtained by collecting terrain elevation values ​​within a specified area using an RTK-GPS receiver. By using a thermal infrared imager, the surface temperature of a designated area is collected to obtain data on the intensity of the urban heat island.

4. The intelligent identification method for rime formation weather based on multi-source meteorological data fusion according to claim 1, characterized in that, S3 includes: S3.1: Perform data cleaning on all sub-data items in the basic dataset by filling in missing values, and normalize all sub-data items in the basic dataset to the range of [0, 1] based on the normalization formula; S3.2: Identify ambient temperature data and precipitation type data. When the ambient temperature data is less than 0 degrees Celsius and the precipitation type data is greater than or equal to 2, obtain cold rain characteristic data with a value of 1. Conversely, if the value is 0, then cold rain characteristic data with a value of 0 are obtained; S3.3: Calculate wind chill characteristic data based on ambient temperature and wind speed data. The specific formula for the calculation is as follows: ； Obtain wind-cold characteristic data ,in, For ambient temperature data, For ambient wind speed data; S3.4: Based on environmental wind speed data and water vapor flux data, the water vapor characteristics are calculated. The specific formula for the calculation is as follows: Water vapor characteristic data were obtained. ,in, For divergence operators, This refers to water vapor flux data; S3.5: Calculate the terrain uplift characteristics based on terrain height data. The specific formula for the calculation is as follows: ； Obtain topographic uplift feature data ,in, To maximize the function, For terrain height data, This is the baseline terrain height data; S3.6: By calculating ambient temperature data and 0.5 times the heat island intensity data The sum of these values ​​yields the corrected feature data. ; S3.7: Subtract the ambient temperature data from the cloud temperature data to obtain the temperature difference characteristic data. ; S3.8: Pack together cold rain feature data, wind chill feature data, water vapor feature data, topographic uplift feature data, correction feature data, and temperature difference feature data to obtain feature vectors.

5. The intelligent identification method for rime formation weather based on multi-source meteorological data fusion according to claim 1, characterized in that, S4 includes: S4.1: Based on the historical feature dataset, the nonlinear hybrid mechanism model is trained. The specific formula for the nonlinear hybrid mechanism model is as follows: ； Obtain the first rime formation value ,in, For characteristic collaborative gating factors, It is a thermodynamic oscillation modulator. The phase transition energy threshold, For the number of historical windows, For the first Each historical weighting factor For the first A topographical historical memory core Activation function, As a dynamic weighting factor, For partial derivatives, For the first Sub-feature data in a feature vector For timestamps, It is the hyperbolic tangent function. For cloud cold energy integrator; Step S4.2: Decompose the computational formulas for the feature co-gating factor, thermodynamic oscillation modulator, phase change energy threshold, terrain history memory kernel, and cloud cold energy integrator in step S4.

1. The specific expression formula set is as follows: ； in, For the Sigmoid function, For the first The mean of the eigenvectors of the terms. For the first The standard deviation of the eigenvectors of the term, For the first A time decay factor, This is the maximum terrain height data. For minimum terrain height data, For the first A historical value for rime formation, This represents the number of spatial grids. Step S4.3: When the nonlinear hybrid mechanism model in step S4.1 has been trained for a preset number of iterations, the rime ice recognition model is output. Step S4.4: Input the feature vector into the rime recognition model and output the rime formation value; Step S4.5: Output the frost formation value to steps S5 and S6.

6. The intelligent identification method for rime formation weather based on multi-source meteorological data fusion according to claim 1, characterized in that, S5 includes: Based on the influence threshold interval (W1, W2), the rime formation value is substituted into the influence threshold interval for comparison; When the rime formation value is less than W1, a no-risk report is generated; when the rime formation value is greater than or equal to W1 and less than W2, a medium-risk report is generated; when the rime formation value is greater than or equal to W2, a high-risk report is generated. The no-risk report includes a statement indicating that the current level of rime ice danger is low, and instructs staff to follow the pre-set operating procedures and ensure continuous monitoring of weather conditions. The medium-risk report includes an explanation of the current level of rime ice danger, requesting staff to activate traffic warning measures and deploy drones to inspect key power line sections; The high-risk report includes an explanation of the current high level of danger from rime ice, requesting staff to issue traffic closure warnings, close bridge sections, and activate DC de-icing devices on power lines; Package the no-risk report, medium-risk report, and high-risk report to obtain the intelligent response report.

7. The intelligent identification method for rime formation weather based on multi-source meteorological data fusion according to claim 1, characterized in that, S6 includes: Based on the rime formation value and feature vector, the contribution rate of key feature data is calculated using the following formula: ； Get the first Contribution rate of feature data ,in, For the first Weighting factors for feature data For the first Feature vector of feature data, This represents the value for rime ice formation. All contribution rates are packaged, grouped in descending order of contribution rate value, and combined with corresponding feature data symbols to obtain an auxiliary identification report.

8. The intelligent identification method for rime formation weather based on multi-source meteorological data fusion according to claim 1, characterized in that, S7 includes: Based on the established manual inspection and observation mechanism, inspectors upload actual rime thickness data and impact range score data via a mobile app; The actual rime thickness data and the impact range score data are packaged together to obtain real-time reporting results.

9. The intelligent identification method for rime formation weather based on multi-source meteorological data fusion according to claim 1, characterized in that, S8 includes: Based on the rime formation value and real-time reporting results; When the absolute value of the rime formation value minus the impact range score data is greater than 2, a weight factor adjustment report is generated; When the actual rime thickness is greater than or equal to 5, a feature weight factor adjustment report is generated; The weighting factor adjustment report includes a set of data representing an increase or decrease of 5% in the weighting factor. The instructions; The feature weighting factor adjustment report includes a set of instructions representing an increase of 3% in the weighting factor for cold rain feature data; Transmit the weight factor adjustment report to step S6, and send the feature weight factor adjustment report to step S4.

1.

10. The intelligent identification method for rime formation weather based on multi-source meteorological data fusion according to claim 1, characterized in that, S9 includes: The system dataset includes rime formation dataset, multi-source depth dataset, feature vectors, rime formation values, intelligent response report, assisted identification report, real-time reporting results, and parameter adjustment report; When the smart response report is a risk-free report, the risk-free report will be sent to the staff's email address via email. When the intelligent response report is a medium-risk report, the medium-risk report will be sent to the staff's email address via email, and the staff will be reminded to check the email via SMS. When the intelligent response report is a high-risk report, the high-risk report will be sent to the staff's email address via email, and the staff will be reminded to check the email via telephone.

Images