A method and system for intelligently revising weather forecasts

Abstract

The present invention discloses an intelligent revision method and system for weather forecast, which relates to the field of weather forecast, and the method comprises: using the optical flow method to analyze a preset radar echo image to generate a first weather forecast; forming a historical radar echo image set, including a plurality of radar echo images with time series characteristics; extracting target radar echo images, and collecting actual meteorological data; extracting a target feature set; forming a single prediction model set; sequentially training to obtain a plurality of target radar extrapolation models; integrating and fusing to obtain a weather forecast model; and intelligently revising the first weather forecast through the weather forecast model to generate a second weather forecast. The present invention solves the problem that the radar echo image extrapolation in the prior art does not make full use of the massive historical radar echo data, resulting in a large deviation in the weather forecast result; by intelligently revising the traditional weather forecast result, the technical effect of improving the accuracy and reliability of the weather forecast is achieved.

Description

Technical Field

[0001] The present invention relates to the field of weather forecasting, and in particular to an intelligent revision method and system for weather forecasting. Background Art

[0002] Weather forecast refers to the analysis and prediction of weather conditions within a specific time range in the future. Accurate and effective weather forecasts are of great value to transportation, agriculture, urban disaster prevention and other fields. Nowcasting mainly refers to the weather forecast for the next 0-2 hours. The core problem of its forecast is to accurately extrapolate radar echo images within a period of time in the future. The radar echo image extrapolation methods in the prior art are mainly optical flow method and cross-correlation method. Whether the optical flow method or the cross-correlation method is used for extrapolation prediction, there is a problem that the massive historical radar echo data is not fully utilized, resulting in a large deviation in the prediction results. For example, in the traditional optical flow method, only two radar echo images, the latest moment and the previous moment, are used for calculation. Therefore, the study of using artificial intelligence and other related technologies to perform data analysis on massive historical radar echo images to revise traditional weather forecast results has important social significance for improving the accuracy and reliability of weather forecasts.

[0003] However, the radar echo image extrapolation in the prior art does not fully utilize the massive historical radar echo data, resulting in large deviations in weather forecast results, which in turn affects the utilization rate of weather forecasts in actual production and life. Summary of the invention

[0004] The purpose of the present invention is to provide an intelligent revision method and system for weather forecasts, so as to solve the problem that the radar echo image extrapolation in the prior art does not make full use of the massive historical radar echo data, resulting in large deviations in weather forecast results, thereby affecting the utilization rate of weather forecasts in actual production and life.

[0005] In view of the above problems, the present invention provides an intelligent revision method and system for weather forecast.

[0006] In a first aspect, the present invention provides a method for intelligently revising a weather forecast, the method comprising: obtaining a preset radar echo image by acquisition, and using an optical flow method to perform extrapolation analysis on the preset radar echo image to generate a first weather forecast; forming a historical radar echo image set, wherein the historical radar echo image set includes multiple radar echo images with time series characteristics; extracting target radar echo images from the multiple radar echo images with time series characteristics, and acquiring actual meteorological data of the target radar echo images; performing multi-dimensional feature extraction on the target radar echo images to obtain a target feature set; forming a single prediction model set based on algorithm principle analysis, wherein the single prediction model set includes multiple single prediction models; using the target feature set and the actual meteorological data as input information, training the multiple single prediction models in turn, and obtaining multiple target radar extrapolation models respectively; based on the principle of the Stacking algorithm, integrating and fusing the multiple target radar extrapolation models to obtain a weather forecast model; intelligently revising the first weather forecast through the weather forecast model, and generating a second weather forecast according to the intelligent revision result.

[0007] In a second aspect, the present invention further provides an intelligent revision system for weather forecasts, which is used to execute an intelligent revision method for weather forecasts as described in the first aspect, wherein the system comprises: a forecast generation model, which is used to acquire a preset radar echo image, and use an optical flow method to perform an extrapolation analysis on the preset radar echo image to generate a first weather forecast; a revision training model, comprising: a first formation model, which is used to form a historical radar echo image set, wherein the historical radar echo image set includes multiple radar echo images with time series characteristics; an acquisition model, which is used to extract a target radar echo image from the multiple radar echo images with time series characteristics, and acquire actual meteorological data of the target radar echo image; a first acquisition model, which is used to The target radar echo image is subjected to multi-dimensional feature extraction to obtain a target feature set; a second model building model is used to build a single prediction model set based on algorithm principle analysis, wherein the single prediction model set includes multiple single prediction models; a training model is used to use the target feature set and the actual meteorological data as input information, and train the multiple single prediction models in turn to obtain multiple target radar extrapolation models respectively; a second acquisition model is used to integrate and fuse the multiple target radar extrapolation models based on the Stacking algorithm principle to obtain a meteorological prediction model; an intelligent revision model is used to intelligently revise the first meteorological forecast through the meteorological prediction model, and generate a second meteorological forecast according to the intelligent revision result.

[0008] One or more technical solutions provided in the present invention have at least the following technical effects or advantages:

[0009] A preset radar echo image is acquired through collection, and the preset radar echo image is extrapolated and analyzed by the optical flow method to generate a first weather forecast; a historical radar echo image set is formed, wherein the historical radar echo image set includes multiple radar echo images with time series characteristics; a target radar echo image is extracted from the multiple radar echo images with time series characteristics, and actual meteorological data of the target radar echo image is acquired; multi-dimensional feature extraction is performed on the target radar echo image to obtain a target feature set; a single prediction model set is formed based on algorithm principle analysis, wherein the single prediction model set includes multiple single prediction models; the target feature set and the actual meteorological data are used as input information, and the multiple single prediction models are trained in turn to obtain multiple target radar extrapolation models respectively; based on the principle of Stacking algorithm, the multiple target radar extrapolation models are integrated and fused to obtain a weather forecast model; the first weather forecast is intelligently revised by the weather forecast model, and a second weather forecast is generated according to the intelligent revision result. By using the traditional optical flow method to analyze the preset radar echo image and obtain the first weather forecast, the preliminary weather forecast is achieved, and the technical goal of providing a revision basis for the subsequent intelligent revision forecast results is provided. Then, by forming a historical radar echo image set, a large amount of historical radar echo data is provided for the data analysis and intelligent revision of the intelligent revision system, and then the target feature set is obtained through feature extraction, and multiple single prediction models are obtained by training the target feature set, achieving the technical goal of providing a single model basis for the subsequent integrated fusion model. Then, the multiple single prediction models obtained by training are integrated through the principle of the Stacking algorithm, and the prediction performance of each integrated model is tested in turn, so as to select a weather forecast model, and finally the weather forecast is made through the weather forecast model with the best prediction performance, and the first weather forecast of the traditional optical flow method is revised according to the weather forecast results of the model, and the second weather forecast is obtained, achieving the technical effect of intelligently revising the traditional weather forecast results and improving the accuracy and reliability of weather forecasts.

[0010] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0012] Figure 1 A schematic diagram of a flow chart of an intelligent revision method for weather forecasts according to the present invention;

[0013] Figure 2 A schematic diagram of a process of adding the plurality of difference feature sets to the target feature set in an intelligent revision method for weather forecasts of the present invention;

[0014] Figure 3 A schematic diagram of a process of using the first historical meteorological element information and the second historical meteorological element information as the actual meteorological data in an intelligent revision method of a meteorological forecast according to the present invention;

[0015] Figure 4 An integrated meteorological prediction model for reversely matching the best prediction result in an intelligent revision method for meteorological forecasts of the present invention, and serving as a flow chart of the meteorological prediction model;

[0016] Figure 5 The present invention is a structural schematic diagram of an intelligent revision system for weather forecast.

[0017] Description of reference numerals:

[0018] Forecast generation model M100, revised training model M200, first assembly model M210, acquisition model M220, first acquisition model M230, second assembly model M240, training model M250, second acquisition model M260, intelligent revision model M300. DETAILED DESCRIPTION

[0019] The present invention provides an intelligent revision method and system for weather forecasts, which solves the problem that radar echo image extrapolation in the prior art does not fully utilize massive historical radar echo data, resulting in large deviations in weather forecast results, thereby affecting the utilization rate of weather forecasts in actual production and life. By intelligently revising traditional weather forecast results, the technical effect of improving the accuracy and reliability of weather forecasts is achieved.

[0020] The acquisition, storage, use, and processing of data in the technical solution of the present invention are in compliance with the relevant provisions of national laws and regulations.

[0021] Below, the technical solutions in the present invention will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments of the present invention. It should be understood that the present invention is not limited to the embodiments described herein. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. It should also be noted that, for the convenience of description, only the parts related to the present invention are shown in the accompanying drawings, rather than all of them.

[0022] Embodiment 1

[0023] Please see attached Figure 1 The present invention provides a method for intelligently revising a weather forecast, wherein the method is applied to an intelligently revising system for a weather forecast, and the method specifically comprises the following steps:

[0024] Step S100: acquiring a preset radar echo image, and using an optical flow method to perform extrapolation analysis on the preset radar echo image to generate a first weather forecast;

[0025] Specifically, the intelligent revision method of weather forecast is applied to an intelligent revision system, and the intelligent revision system can achieve the effect of improving the accuracy and reliability of weather forecast by intelligently revising traditional weather forecast results.

[0026] First, the traditional optical flow method is used to perform extrapolation analysis on the preset radar echo image, and the first weather forecast is generated according to the extrapolation analysis results. Among them, the optical flow can be considered as the velocity field of the pixels in the image, that is, the motion field composed of the instantaneous change rate. By applying the optical flow to the known image at the current moment, the predicted image can be obtained. The traditional optical flow method is based on three major assumptions, including: the first assumption that in two adjacent frames of images, the brightness of all pixels does not change but only shifts, that is, the brightness is constant; the second assumption that the difference between the two adjacent frames of images is not large, and the motion displacement of the corresponding pixels is very small, that is, small motion; the third assumption that the motion patterns of adjacent pixels in the same image are similar, that is, spatial consistency. Since the three major assumptions of the optical flow method are not valid in the actual radar echo image, and the optical flow method only uses the radar data at the current moment, and does not effectively use a large amount of existing radar echo historical data, the generated first weather forecast has a large deviation. However, the first weather forecast is obtained by analyzing the optical flow method, which realizes the technical goal of preliminary weather forecast and provides a revision basis for the subsequent intelligent revision of the forecast results.

[0027] Step S200: assembling a historical radar echo image set, wherein the historical radar echo image set includes a plurality of radar echo images with time series characteristics;

[0028] Step S300: extracting a target radar echo image from the plurality of radar echo images with time series characteristics, and collecting actual meteorological data of the target radar echo image;

[0029] Further, as attached Figure 3 As shown, step S300 of the present invention specifically includes:

[0030] Step S310: collecting target historical meteorological data records according to the target time series characteristics, wherein the target historical meteorological data records refer to historical meteorological data under the target time series characteristics;

[0031] Step S320: forming a meteorological element set, wherein the meteorological element set includes a first meteorological element and a second meteorological element;

[0032] Step S330: extracting element information from the target historical meteorological data record based on the first meteorological element and the second meteorological element, and obtaining first historical meteorological element information and second historical meteorological element information respectively;

[0033] Step S340: taking the first historical meteorological element information and the second historical meteorological element information as the actual meteorological data.

[0034] Specifically, before constructing an intelligent model for intelligently revising the first weather forecast, a historical radar echo image set is first automatically collected and assembled, wherein the historical radar echo image set includes multiple radar echo images with time series characteristics. Then, a target radar echo image is extracted from the multiple radar echo images with time series characteristics, and after feature collection and analysis of the target radar echo image, intelligent training of the revision model is performed. The target radar echo image refers to any one of the multiple radar echo images with time series characteristics.

[0035] Furthermore, according to the target time series characteristics of the target radar echo image, the historical meteorological data of the same period is retrieved, that is, the target historical meteorological data record is obtained. Then, a comprehensive analysis is performed and a meteorological element set is formed, such as exemplary meteorological elements such as temperature, air pressure, wind, humidity, cloud, precipitation and various weather phenomena. Then, based on each meteorological element in the meteorological element set, the target historical meteorological data record is traversed and matched in turn, so as to obtain the actual meteorological data corresponding to each meteorological element in the meteorological element set, that is, the first historical meteorological element information and the second historical meteorological element information are obtained. Finally, the result after the first historical meteorological element information and the second historical meteorological element information are merged is used as the actual meteorological data.

[0036] Step S400: performing multi-dimensional feature extraction on the target radar echo image to obtain a target feature set;

[0037] Further, as attached Figure 2 As shown, step S400 of the present invention further includes:

[0038] Step S410: performing image difference analysis based on the target radar echo image and combining the multiple radar echo images with time series characteristics to obtain multiple difference feature sets;

[0039] Step S420: Add the multiple difference feature sets to the target feature set.

[0040] Furthermore, step S410 includes the following steps:

[0041] Step S411: obtaining target time series characteristics of the target radar echo image;

[0042] Step S412: extracting a first radar echo image from the plurality of radar echo images having the time series characteristics based on the target time series characteristics;

[0043] Step S413: wherein the first radar echo image and the target radar echo image have a first continuity;

[0044] Step S414: performing image difference analysis on the target radar echo image and the first radar echo image to obtain a first difference feature set;

[0045] Step S415: obtaining a first time series feature of the first radar echo image;

[0046] Step S416: extracting a second radar echo image from the plurality of radar echo images having the time series feature based on the first time series feature;

[0047] Step S417: wherein the second radar echo image and the target radar echo image have a second continuity;

[0048] Step S418: performing image difference analysis on the target radar echo image and the second radar echo image to obtain a second difference feature set;

[0049] Step S419: assembling the plurality of difference feature sets according to the first difference feature set and the second difference feature set.

[0050] Specifically, firstly, the target radar echo image is subjected to multi-dimensional feature extraction, such as the spatiotemporal features in the image, the overall morphological features of the object, the motion trend and the detail changes, and the target feature set is formed. Then, the target radar echo image is sequentially compared with the related radar echo images, and a plurality of difference feature sets are obtained according to the comparison results, and then the plurality of difference feature sets obtained by the comparison analysis are added to the target feature set for the training basis of the subsequent model.

[0051] Specifically, firstly, the target time series features of the target radar echo image are analyzed and determined, and based on the target time series features, the first radar echo image is extracted from the multiple radar echo images with time series features. The first radar echo image refers to the first radar echo image obtained before the target radar echo image is scanned and traversed among the multiple radar echo images with time series features, that is, there is a first continuity between the first radar echo image and the target radar echo image, and the first continuity means that the first radar echo image is obtained first and then the target radar echo image is obtained. Then, the target radar echo image and the first radar echo image are subjected to image difference analysis to obtain a first difference feature set. That is to say, as the time series changes, the target radar echo image is a set of change features based on the first radar echo image. Further, the first time series feature of the first radar echo image is analyzed and determined, and based on the first time series feature, a second radar echo image is extracted from the multiple radar echo images with time series features, wherein the second radar echo image has a second continuity with the target radar echo image, and the target radar echo image and the second radar echo image are subjected to image difference analysis to obtain a second difference feature set. The iteration continues to the first radar echo image of the multiple radar echo images with time series features, and the change features of each radar echo image are collected, that is, according to the first difference feature set and the second difference feature set, the plurality of difference feature sets are obtained by merging.

[0052] By forming a historical radar echo image set, a large amount of historical radar echo data is provided for the data analysis and intelligent revision of the intelligent revision system, and then the target feature set is obtained through feature extraction. At the same time, multiple difference feature sets are obtained through image comparison and analysis, achieving the technical goal of providing a single model basis for the subsequent integrated fusion model.

[0053] Step S500: Analyze and establish a single prediction model set based on the algorithm principle, wherein the single prediction model set includes multiple single prediction models;

[0054] Step S600: using the target feature set and the actual meteorological data as input information, training the multiple single prediction models in sequence to obtain multiple target radar extrapolation models respectively;

[0055] Specifically, a single prediction model set is formed based on the analysis of algorithm principles, wherein the single prediction model set includes multiple single prediction models. For example, three different learning and training mechanisms are selected, and multiple single prediction models are trained in sequence, such as a support vector machine correction model, a recurrent neural network correction model, and a gradient boosting decision tree correction model. Then, the target feature set and the actual meteorological data are used as model training data, and the multiple single prediction models are input in sequence and trained separately, and finally multiple target radar extrapolation models are obtained. Multiple target radar extrapolation models are obtained through training, which provide a model basis for comprehensive and multi-angle meteorological forecast revisions, and ultimately improve the quality of meteorological forecast revisions by learning from each other's performance.

[0056] Step S700: Based on the principle of Stacking algorithm, the multiple target radar extrapolation models are integrated and fused to obtain a weather forecast model;

[0057] Further, as attached Figure 4 As shown, step S700 of the present invention specifically includes:

[0058] Step S710: Based on the principle of Stacking algorithm, the multiple target radar extrapolation models are integrated and fused to obtain multiple integrated meteorological prediction models;

[0059] Step S720: constructing an integrated model data set based on the target feature set and the actual meteorological data;

[0060] Step S730: predicting the integrated model data set using the multiple integrated meteorological prediction models in sequence to obtain multiple prediction results;

[0061] Further, step S730 includes the following steps:

[0062] Step S731: extracting any one integrated meteorological prediction model from the plurality of integrated meteorological prediction models;

[0063] Step S732: obtaining a primary learner and a meta learner of any one of the integrated meteorological forecast models;

[0064] Step S733: predicting the integrated model data set by the primary learner to obtain a primary prediction result;

[0065] Step S734: using the primary prediction result as input information of the meta-learner, and processing to obtain output information of the meta-learner;

[0066] Step S735: Based on the output information of the meta-learner, obtain the prediction result of any one of the integrated meteorological prediction models.

[0067] Step S740: Filtering the best prediction result among the multiple prediction results;

[0068] Further, step S740 includes the following steps:

[0069] Step S741: extracting the first prediction result from the multiple prediction results;

[0070] Step S742: wherein the first prediction result includes a first meteorological element prediction result and a second meteorological element prediction result;

[0071] Step S743: Compare the first meteorological element prediction result with the first historical meteorological element information to obtain a first comparison result;

[0072] Step S744: comparing the second meteorological element prediction result with the second historical meteorological element information to obtain a second comparison result;

[0073] Step S745: Calculate a first prediction hit rate according to the first comparison result and the second comparison result;

[0074] Step S746: Determine the best prediction result based on the first prediction hit rate.

[0075] Step S750: reverse matching the integrated meteorological prediction model of the best prediction result and using it as the meteorological prediction model.

[0076] Step S800: Intelligently revise the first weather forecast using the weather prediction model, and generate a second weather forecast based on the intelligent revision result.

[0077] Specifically, in order to achieve comprehensive analysis and research of data from multiple aspects and angles, and to complement the performance of different models, ensemble learning methods came into being. Ensemble learning is to improve the effect of machine learning model training by combining multiple models. Compared with a single model, this method can produce better prediction performance.

[0078] First, based on the principle of Stacking algorithm, the multiple target radar extrapolation models are integrated and fused to obtain multiple integrated meteorological prediction models. That is to say, one of the multiple target radar extrapolation models is used as the meta-learner of the second layer of the Stacking framework, and the other target radar extrapolation models are used as the initial learners of the first layer of the Stacking framework, and finally multiple integrated meteorological prediction models are formed. Then, the target feature set and the actual meteorological data are formed into an integrated model data set for training and predicting each integrated meteorological prediction model. Further, the integrated model data set is predicted by the multiple integrated meteorological prediction models in turn to obtain multiple prediction results. Specifically, first, any one of the multiple integrated meteorological prediction models is extracted to obtain the primary learner and meta-learner of any one of the integrated meteorological prediction models. Then, the integrated model data set is predicted by the primary learner to obtain the primary prediction result, and the primary prediction result is used as the input information of the meta-learner to process and obtain the output information of the meta-learner. Finally, based on the output information of the meta-learner, the prediction result of any one of the integrated meteorological prediction models is obtained. The multiple prediction results are obtained by training and predicting the multiple integrated meteorological prediction models in sequence.

[0079] Further, after obtaining the multiple prediction results, in order to objectively evaluate the prediction results of each model, the hit rate of each model prediction result is calculated in turn. First, the first prediction result among the multiple prediction results is extracted. Among them, the first prediction result includes the prediction result of the first meteorological element and the prediction result of the second meteorological element. Then the first meteorological element prediction result is compared with the first historical meteorological element information, that is, the meteorological element data predicted by the model is compared with the actual meteorological element data in the historical record, that is, the first comparison result is obtained. For example, the prediction deviation is calculated and the deviation level to which the deviation degree belongs is determined. Then, the second meteorological element prediction result is compared with the second historical meteorological element information to obtain the second comparison result. Finally, according to the first comparison result and the second comparison result, the first prediction hit rate of the first prediction result is calculated. The hit rates of the multiple prediction results are compared in turn, and the best prediction result among the multiple prediction results is selected, that is, the prediction result with the highest hit rate is used as the best prediction result. Next, the integrated meteorological prediction model of the best prediction result is reversely matched and used as the meteorological prediction model. Finally, the preset radar echo image is predicted by the weather prediction model to obtain the model prediction result, the first weather forecast determined by the optical flow method is intelligently revised by the model prediction result, and the second weather forecast is generated according to the intelligent revision result. By intelligently revising the traditional weather forecast results, the technical effect of improving the accuracy and reliability of weather forecasts is achieved.

[0080] In summary, the intelligent revision method of weather forecast provided by the present invention has the following technical effects:

[0081] A preset radar echo image is acquired through collection, and the preset radar echo image is extrapolated and analyzed by the optical flow method to generate a first weather forecast; a historical radar echo image set is formed, wherein the historical radar echo image set includes multiple radar echo images with time series characteristics; a target radar echo image is extracted from the multiple radar echo images with time series characteristics, and actual meteorological data of the target radar echo image is acquired; multi-dimensional feature extraction is performed on the target radar echo image to obtain a target feature set; a single prediction model set is formed based on algorithm principle analysis, wherein the single prediction model set includes multiple single prediction models; the target feature set and the actual meteorological data are used as input information, and the multiple single prediction models are trained in turn to obtain multiple target radar extrapolation models respectively; based on the principle of Stacking algorithm, the multiple target radar extrapolation models are integrated and fused to obtain a weather forecast model; the first weather forecast is intelligently revised by the weather forecast model, and a second weather forecast is generated according to the intelligent revision result. By using the traditional optical flow method to analyze the preset radar echo image and obtain the first weather forecast, the preliminary weather forecast is achieved, and the technical goal of providing a revision basis for the subsequent intelligent revision forecast results is provided. Then, by forming a historical radar echo image set, a large amount of historical radar echo data is provided for the data analysis and intelligent revision of the intelligent revision system, and then the target feature set is obtained through feature extraction, and multiple single prediction models are obtained by training the target feature set, achieving the technical goal of providing a single model basis for the subsequent integrated fusion model. Then, the multiple single prediction models obtained by training are integrated through the principle of the Stacking algorithm, and the prediction performance of each integrated model is tested in turn, so as to select a weather forecast model, and finally the weather forecast is made through the weather forecast model with the best prediction performance, and the first weather forecast of the traditional optical flow method is revised according to the weather forecast results of the model, and the second weather forecast is obtained, achieving the technical effect of intelligently revising the traditional weather forecast results and improving the accuracy and reliability of weather forecasts.

[0082] Embodiment 2

[0083] Based on the same inventive concept as the intelligent revision method of a weather forecast in the aforementioned embodiment, the present invention also provides an intelligent revision system for a weather forecast, see the attached Figure 5 , the system comprising:

[0084] A forecast generation model M100, wherein the forecast generation model M100 is used to acquire a preset radar echo image and perform an extrapolation analysis on the preset radar echo image using an optical flow method to generate a first weather forecast;

[0085] The revised training model M200 includes:

[0086] A first assembly model M210, wherein the first assembly model M210 is used to assemble a historical radar echo image set, wherein the historical radar echo image set includes a plurality of radar echo images with time series characteristics;

[0087] An acquisition model M220, wherein the acquisition model M220 is used to extract a target radar echo image from the plurality of radar echo images with time series characteristics, and to acquire actual meteorological data of the target radar echo image;

[0088] A first acquisition model M230, wherein the first acquisition model M230 is used to perform multi-dimensional feature extraction on the target radar echo image to obtain a target feature set;

[0089] A second building model M240, wherein the second building model M240 is used to build a single prediction model set based on algorithm principle analysis, wherein the single prediction model set includes multiple single prediction models;

[0090] A training model M250, wherein the training model M250 is used to take the target feature set and the actual meteorological data as input information, train the multiple single prediction models in sequence, and obtain multiple target radar extrapolation models respectively;

[0091] A second acquisition model M260, wherein the second acquisition model M260 is used to integrate and fuse the multiple target radar extrapolation models based on the Stacking algorithm principle to obtain a meteorological prediction model;

[0092] The intelligent revision model M300 is used to intelligently revise the first weather forecast through the weather prediction model, and generate a second weather forecast according to the intelligent revision result.

[0093] Further, the first acquisition model M230 in the system is also used for:

[0094] Based on the target radar echo image, image difference analysis is performed in combination with the multiple radar echo images with time series characteristics to obtain multiple difference feature sets;

[0095] The plurality of difference feature sets are added to the target feature set.

[0096] Further, the first acquisition model M230 in the system is also used for:

[0097] Obtaining target time series characteristics of the target radar echo image;

[0098] Extracting a first radar echo image from the plurality of radar echo images having the time series characteristics based on the target time series characteristics;

[0099] Wherein, the first radar echo image and the target radar echo image have a first continuity;

[0100] Performing image difference analysis on the target radar echo image and the first radar echo image to obtain a first difference feature set;

[0101] Obtaining a first time series feature of the first radar echo image;

[0102] Extracting a second radar echo image from the plurality of radar echo images having the time series feature based on the first time series feature;

[0103] Wherein, the second radar echo image and the target radar echo image have a second continuity;

[0104] Performing image difference analysis on the target radar echo image and the second radar echo image to obtain a second difference feature set;

[0105] The multiple difference feature sets are assembled based on the first difference feature set and the second difference feature set.

[0106] Furthermore, the acquisition model M220 in the system is also used for:

[0107] Collecting target historical meteorological data records according to the target time series characteristics, wherein the target historical meteorological data records refer to historical meteorological data under the target time series characteristics;

[0108] Establishing a meteorological element set, wherein the meteorological element set includes a first meteorological element and a second meteorological element;

[0109] Based on the first meteorological element and the second meteorological element, extracting element information from the target historical meteorological data record to obtain first historical meteorological element information and second historical meteorological element information respectively;

[0110] The first historical meteorological element information and the second historical meteorological element information are used as the actual meteorological data.

[0111] Further, the second acquisition model M260 in the system is also used for:

[0112] Based on the principle of Stacking algorithm, the multiple target radar extrapolation models are integrated and fused to obtain multiple integrated meteorological prediction models;

[0113] Based on the target feature set and the actual meteorological data, an integrated model data set is formed;

[0114] Predicting the integrated model data set using the multiple integrated meteorological prediction models in sequence to obtain multiple prediction results;

[0115] Filtering the best prediction result among the multiple prediction results;

[0116] The integrated meteorological prediction model of the best prediction result is reversely matched and used as the meteorological prediction model.

[0117] Further, the second acquisition model M260 in the system is also used for:

[0118] Extracting any one integrated meteorological prediction model from the plurality of integrated meteorological prediction models;

[0119] Obtaining a primary learner and a meta-learner of any one of the integrated meteorological forecast models;

[0120] Predicting the integrated model data set by the primary learner to obtain a primary prediction result;

[0121] Using the primary prediction result as input information of the meta-learner, and processing to obtain output information of the meta-learner;

[0122] Based on the output information of the meta-learner, a prediction result of any one of the integrated meteorological prediction models is obtained.

[0123] Further, the second acquisition model M260 in the system is also used for:

[0124] Extracting a first prediction result from the plurality of prediction results;

[0125] Wherein, the first prediction result includes a first meteorological element prediction result and a second meteorological element prediction result;

[0126] Comparing the first meteorological element prediction result with the first historical meteorological element information to obtain a first comparison result;

[0127] Comparing the second meteorological element prediction result with the second historical meteorological element information to obtain a second comparison result;

[0128] Calculating a first prediction hit rate according to the first comparison result and the second comparison result;

[0129] The best prediction result is determined based on the first prediction hit rate.

[0130] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. Figure 1The intelligent revision method and specific embodiments of a weather forecast in the first embodiment are also applicable to the intelligent revision system of a weather forecast in the present embodiment. Through the above detailed description of the intelligent revision method of a weather forecast, those skilled in the art can clearly know the intelligent revision system of a weather forecast in the present embodiment, so for the sake of brevity of the specification, it will not be described in detail here. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0131] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0132] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention belong to the scope of the present invention and its equivalent technologies, the present invention is also intended to include these changes and variations.

Claims

1. An intelligent revision method for weather forecast, characterized in that: include: Acquire a preset radar echo image, and use an optical flow method to perform extrapolation analysis on the preset radar echo image to generate a first weather forecast; Establishing a historical radar echo image set, wherein the historical radar echo image set includes a plurality of radar echo images with time series characteristics; Extracting a target radar echo image from the plurality of radar echo images with time series characteristics, and collecting actual meteorological data of the target radar echo image; Performing multi-dimensional feature extraction on the target radar echo image to obtain a target feature set; Analyzing and forming a single prediction model set based on algorithm principles, wherein the single prediction model set includes multiple single prediction models; Taking the target feature set and the actual meteorological data as input information, training the multiple single prediction models in sequence to obtain multiple target radar extrapolation models respectively; Based on the principle of Stacking algorithm, the multiple target radar extrapolation models are integrated and fused to obtain a weather forecast model; The first weather forecast is intelligently revised using the weather prediction model, and a second weather forecast is generated based on the intelligent revision result.

2. The intelligent revision method according to claim 1, characterized in that: After extracting multi-dimensional features from the target radar echo image to obtain a target feature set, the method further includes: Based on the target radar echo image, image difference analysis is performed in combination with the multiple radar echo images with time series characteristics to obtain multiple difference feature sets; The plurality of difference feature sets are added to the target feature set.

3. The intelligent revision method according to claim 2, characterized in that: The image difference analysis is performed based on the target radar echo image and in combination with the multiple radar echo images with time series characteristics to obtain multiple difference feature sets, including: Obtaining target time series characteristics of the target radar echo image; Extracting a first radar echo image from the plurality of radar echo images having the time series characteristics based on the target time series characteristics; Wherein, the first radar echo image and the target radar echo image have a first continuity; Performing image difference analysis on the target radar echo image and the first radar echo image to obtain a first difference feature set; Obtaining a first time series feature of the first radar echo image; Extracting a second radar echo image from the plurality of radar echo images having the time series feature based on the first time series feature; Wherein, the second radar echo image and the target radar echo image have a second continuity; Performing image difference analysis on the target radar echo image and the second radar echo image to obtain a second difference feature set; The multiple difference feature sets are assembled based on the first difference feature set and the second difference feature set.

4. The intelligent revision method according to claim 1, characterized in that: The step of extracting a target radar echo image from the plurality of radar echo images having time series characteristics and collecting actual meteorological data of the target radar echo image comprises: Obtaining target time series characteristics of the target radar echo image; collecting target historical meteorological data records according to the target time series characteristics, wherein the target historical meteorological data records refer to historical meteorological data under the target time series characteristics; Establishing a meteorological element set, wherein the meteorological element set includes a first meteorological element and a second meteorological element; Based on the first meteorological element and the second meteorological element, extracting element information from the target historical meteorological data record to obtain first historical meteorological element information and second historical meteorological element information respectively; The first historical meteorological element information and the second historical meteorological element information are used as the actual meteorological data.

5. The intelligent revision method according to claim 4, characterized in that: Based on the principle of Stacking algorithm, the multiple target radar extrapolation models are integrated and fused to obtain a weather forecast model, including: Based on the principle of Stacking algorithm, the multiple target radar extrapolation models are integrated and fused to obtain multiple integrated meteorological prediction models; Based on the target feature set and the actual meteorological data, an integrated model data set is formed; Predicting the integrated model data set using the multiple integrated meteorological prediction models in sequence to obtain multiple prediction results; Filtering the best prediction result among the multiple prediction results; The integrated meteorological prediction model of the best prediction result is reversely matched and used as the meteorological prediction model.

6. The intelligent revision method according to claim 5, characterized in that: The integrated model data set is predicted in sequence by the multiple integrated meteorological prediction models to obtain multiple prediction results, including: Extracting any one integrated meteorological prediction model from the plurality of integrated meteorological prediction models; Obtaining a primary learner and a meta-learner of any one of the integrated meteorological forecast models; Predicting the integrated model data set by the primary learner to obtain a primary prediction result; Using the primary prediction result as input information of the meta-learner, and processing to obtain output information of the meta-learner; Based on the output information of the meta-learner, a prediction result of any one of the integrated meteorological prediction models is obtained.

7. The intelligent revision method according to claim 6, characterized in that: The screening of the best prediction result among the multiple prediction results comprises: Extracting a first prediction result from the plurality of prediction results; Wherein, the first prediction result includes a first meteorological element prediction result and a second meteorological element prediction result; Comparing the first meteorological element prediction result with the first historical meteorological element information to obtain a first comparison result; Comparing the second meteorological element prediction result with the second historical meteorological element information to obtain a second comparison result; Calculating a first prediction hit rate according to the first comparison result and the second comparison result; The best prediction result is determined based on the first prediction hit rate.

8. An intelligent revision system for weather forecasts, characterized in that: include: A forecast generation model, which is used to collect a preset radar echo image and use an optical flow method to perform extrapolation analysis on the preset radar echo image to generate a first weather forecast; Revising the training model, the revised training model comprising: A first building model, which is used to build a historical radar echo image set, wherein the historical radar echo image set includes a plurality of radar echo images with time series characteristics; An acquisition model, which is used to extract a target radar echo image from the plurality of radar echo images with time series characteristics, and to acquire actual meteorological data of the target radar echo image; First, a model is obtained, which is used to extract multi-dimensional features from the target radar echo image to obtain a target feature set; A second model building unit is used to build a single prediction model set based on algorithm principle analysis, wherein the single prediction model set includes multiple single prediction models; A training model, which is used to take the target feature set and the actual meteorological data as input information, train the multiple single prediction models in sequence, and obtain multiple target radar extrapolation models respectively; Second, a model is obtained, which is used to integrate and fuse the multiple target radar extrapolation models based on the principle of Stacking algorithm to obtain a meteorological prediction model; An intelligent revision model is used to intelligently revise the first weather forecast through the weather prediction model, and generate a second weather forecast according to the intelligent revision result.

9. The intelligent revision system for weather forecast according to claim 8, characterized in that: The acquisition model specifically includes: Obtaining target time series characteristics of the target radar echo image; collecting target historical meteorological data records according to the target time series characteristics, wherein the target historical meteorological data records refer to historical meteorological data under the target time series characteristics; Establishing a meteorological element set, wherein the meteorological element set includes a first meteorological element and a second meteorological element; Based on the first meteorological element and the second meteorological element, extracting element information from the target historical meteorological data record to obtain first historical meteorological element information and second historical meteorological element information respectively; The first historical meteorological element information and the second historical meteorological element information are used as the actual meteorological data.

10. The intelligent revision system for weather forecast according to claim 9, characterized in that: The second acquisition model specifically includes: Based on the principle of Stacking algorithm, the multiple target radar extrapolation models are integrated and fused to obtain multiple integrated meteorological prediction models; Based on the target feature set and the actual meteorological data, an integrated model data set is formed; Predicting the integrated model data set using the multiple integrated meteorological prediction models in sequence to obtain multiple prediction results; Filtering the best prediction result among the multiple prediction results; The integrated meteorological prediction model of the best prediction result is reversely matched and used as the meteorological prediction model.

Images