A low-temperature snow and ice disaster risk prediction method and system based on big data

Abstract

The application relates to a low-temperature rain-snow-ice disaster risk prediction method and system based on big data, which comprises the following steps: collecting multi-source heterogeneous data and preprocessing to obtain a standardized data set, wherein the multi-source heterogeneous data comprises meteorological data, traffic data and power data; according to the standardized data set, dynamic characteristics in the meteorological data are extracted, the dynamic characteristics are filtered, and an optimized feature set is obtained; feature extraction is performed on the optimized feature set to obtain a space-time feature vector; the traffic data and the power data are fused to obtain a comprehensive influence factor; the space-time feature vector and the comprehensive influence factor are combined to obtain a disaster risk feature matrix; and the disaster risk feature matrix is input into a risk prediction model to output a disaster risk grade and a probability distribution, wherein the risk prediction model is constructed based on a long short-term memory network. The application realizes accurate risk prediction of a low-temperature rain-snow-ice disaster scenario, and provides strong decision support for disaster prevention and reduction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information processing, in particular to a low-temperature snow and ice disaster risk prediction method and system based on big data. BACKGROUND

[0002] Low-temperature snow and ice disaster is an important natural disaster that affects social economy and public safety. Its prediction and evaluation are of key significance to reducing losses and ensuring people's livelihood. Traditional methods mainly rely on single data source and simple statistical model, which is difficult to integrate multi-source data and has insufficient prediction accuracy to cope with the dynamic changes of complex meteorological and geographical conditions. This makes real-time evaluation and accurate early warning of disaster risk become a difficult problem to be solved.

[0003] In the field of disaster prediction, the core challenge is how to effectively integrate multi-source heterogeneous data and extract key features from it. The format and dimension of meteorological, traffic, power, geographical and other data are quite different, and traditional processing methods are difficult to unify, resulting in low information fusion efficiency. The difficulty of data integration further affects the accuracy of feature extraction, especially the dynamic features directly related to disaster risk, such as temperature drop rate, snow and ice conversion coefficient, etc., which are difficult to accurately capture by a single algorithm. The extraction quality of these features directly determines the performance of the prediction model. If it cannot be accurately extracted, the model will be difficult to reflect the spatio-temporal evolution law of the disaster, and thus lead to distortion of risk level prediction.

[0004] Therefore, how to efficiently integrate multi-source heterogeneous data, accurately extract dynamic features related to low-temperature snow and ice disaster, and build a high-precision risk prediction model based on it has become a key research problem. SUMMARY

[0005] The purpose of the present application is to provide a low-temperature snow and ice disaster risk prediction method and system based on big data, which realizes accurate risk prediction of low-temperature snow and ice disaster scenarios and provides strong decision support for disaster prevention and reduction.

[0006] To achieve the above purpose, the present application provides the following scheme:

[0007] A low-temperature snow and ice disaster risk prediction method based on big data, comprising:

[0008] Collecting multi-source heterogeneous data and preprocessing to obtain a standardized data set, wherein the multi-source heterogeneous data includes meteorological data, traffic data and power data;

[0009] According to the standardized data set, extracting dynamic features in the meteorological data, screening the dynamic features, and obtaining an optimized feature set;

[0010] Feature extraction is performed on the optimized feature set to obtain a spatio-temporal feature vector;

[0011] Fusing the traffic data and the power data, an integrated influence factor is obtained;

[0012] Combining the spatio-temporal feature vector and the integrated influence factor, a disaster risk feature matrix is obtained;

[0013] Inputting the disaster risk feature matrix into a risk prediction model, a disaster risk grade and a probability distribution are output, wherein the risk prediction model is constructed based on a long short-term memory network and obtained through a training set, and the training set includes historical disaster risk feature matrices.

[0014] Optionally, multi-source heterogeneous data is collected and preprocessed to obtain a standardized data set, including:

[0015] Multi-source heterogeneous data is obtained from meteorological, traffic and power data sources through a preset protocol, and a standard interface and protocol are used for parsing to obtain an original data set;

[0016] If the original data set has missing values or abnormal values, mean filling and K-neighbor algorithm-based cleaning are used to obtain a cleaned data set;

[0017] According to the cleaned data set, a format conversion algorithm is used to map the formats of different data sources into a unified structure to obtain a data set with consistent formats;

[0018] If the data set with consistent formats has inconsistent dimensions, a normalization algorithm is used to standardize the dimensions of numerical data to obtain a data set with consistent dimensions;

[0019] Through the data set with consistent dimensions, data verification rules are used to judge data integrity and consistency to obtain the standardized data set.

[0020] Optionally, according to the standardized data set, dynamic features in the meteorological data are extracted, including: through a principal component analysis algorithm, the standardized data set is processed for dimension reduction, principal components are extracted, and dynamic features including air temperature drop rate and snowfall conversion coefficient are obtained.

[0021] Optionally, the dynamic features are screened to obtain an optimized feature set, including:

[0022] If the cumulative variance contribution rate of the principal components reaches a preset threshold, the corresponding principal components are retained to obtain a preliminary feature set;

[0023] A clustering analysis algorithm is used to group the preliminary feature set to judge the correlation between features to obtain a feature grouping result;

[0024] Select features that achieve a preset correlation degree with the air temperature drop rate and the rain-snow conversion coefficient from the feature grouping result, and construct the optimized feature set.

[0025] Optionally, feature extraction is performed on the optimized feature set to obtain a spatio-temporal feature vector, including:

[0026] The optimized feature set is input into a convolutional neural network, and spatio-temporal feature extraction is performed on the optimized feature set to obtain initial spatio-temporal features.

[0027] If the resolution of the initial spatio-temporal features is lower than a preset threshold, the resolution of the feature map is enhanced through an upsampling operation to obtain an enhanced feature map.

[0028] A pooling layer of the convolutional neural network is used to extract key spatio-temporal features from the enhanced feature map to generate the spatio-temporal feature vector.

[0029] Optionally, the traffic data and the power data are fused to obtain a comprehensive influence factor, including:

[0030] Key traffic features and power grid load area distribution data are extracted from the traffic data and the power data, respectively, wherein the area distribution data includes real-time power fluctuations of different power supply nodes.

[0031] The key traffic features and the power grid load area distribution data are fused by a weighted average method to obtain the comprehensive influence factor.

[0032] Optionally, extracting key traffic features and power grid load area distribution data includes inputting the traffic data and the power data distribution into a random forest algorithm to extract the key traffic features and the power grid load area distribution data.

[0033] Optionally, after outputting the disaster risk level and the probability distribution, if the shortest distance between one region and another region is less than a preset threshold, the regions are merged into the same snow and ice freezing process.

[0034] If the coincidence rate of the meteorological station sites of the snow and ice freezing region on the current day and the snow and ice freezing region of the previous day is greater than a preset value, the two days are the same snow and ice freezing event, otherwise they are two separate snow and ice freezing events.

[0035] The application also provides a system implemented according to a low-temperature snow and ice freezing disaster risk prediction method based on big data, including a data acquisition module, a data processing module, a first fusion module, a second fusion module, and a risk prediction module.

[0036] The data acquisition module is configured to collect and preprocess multi-source heterogeneous data to obtain a standardized data set, wherein the multi-source heterogeneous data includes meteorological data, traffic data, and power data.

[0037] The data processing module is configured to extract dynamic features from the meteorological data, filter the dynamic features, obtain an optimized feature set, and extract features from the optimized feature set to obtain a spatio-temporal feature vector;

[0038] The first fusion module is configured to fuse the traffic data and the power data to obtain a comprehensive influence factor;

[0039] The second fusion module is configured to combine the spatio-temporal feature vector and the comprehensive influence factor to obtain a disaster risk feature matrix;

[0040] The risk prediction module is configured to input the disaster risk feature matrix into a risk prediction model to output a disaster risk level and a probability distribution, wherein the risk prediction model is constructed based on a long short-term memory network and is obtained by training a training set, and the training set includes a historical disaster risk feature matrix.

[0041] The present application has the following advantages: the present application obtains multi-source heterogeneous data such as meteorological data, traffic data, and power data through a preset standardization protocol, and uses a data cleaning and format conversion algorithm to unify the data format. Key features are extracted using principal component analysis. Key traffic features and power grid load area distribution features are analyzed using a random forest algorithm, and a disaster risk feature matrix is obtained by weighted fusion. Finally, a risk prediction model is constructed based on a long short-term memory network, and the model parameters are optimized using a gradient boosting algorithm. The present application realizes accurate risk prediction of low-temperature snow and ice disaster scenarios, and provides strong decision support for disaster prevention and mitigation. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0043] Figure 1 A flowchart of a low-temperature snow and ice disaster risk prediction method based on big data according to an embodiment of the present application. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0045] In order to make the above objectives, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0046] Embodiment one:

[0047] As Figure 1 shown, the present embodiment provides a low-temperature snow and ice disaster risk prediction method based on big data, comprising:

[0048] Collecting multi-source heterogeneous data and preprocessing to obtain a standardized data set, wherein the multi-source heterogeneous data includes meteorological data, traffic data and power data;

[0049] According to the standardized data set, the dynamic features in the meteorological data are extracted, the dynamic features are screened, and an optimized feature set is obtained;

[0050] Feature extraction is performed on the optimized feature set to obtain a spatio-temporal feature vector;

[0051] Traffic data and power data are fused to obtain a comprehensive influence factor;

[0052] The spatio-temporal feature vector and the comprehensive influence factor are combined to obtain a disaster risk feature matrix;

[0053] The disaster risk feature matrix is input into a risk prediction model to output a disaster risk level and a probability distribution, wherein the risk prediction model is constructed based on a long short-term memory network and obtained by training a training set, and the training set includes historical disaster risk feature matrices.

[0054] Specifically, meteorological, traffic and power data are of great significance to the prediction, evaluation and prevention of low-temperature snow and ice disaster risk. By integrating and analyzing these data, the occurrence of disasters can be more accurately predicted, the intensity and influence range of disasters can be evaluated, and scientific decision-making basis can be provided for relevant departments and the public, thereby reducing the possible losses caused by disasters.

[0055] In the present embodiment, the spatio-temporal feature vector and the comprehensive influence factor are combined by a weighted fusion algorithm. Assuming that the traffic flow standardized value at a certain moment is 0.8, the power grid load is 0.6, and the comprehensive influence factor is 0.75, the weighted calculation is (0.7x0.8+0.3x0.6=0.74).

[0056] Further, collecting multi-source heterogeneous data and preprocessing to obtain a standardized data set includes:

[0057] Multi-source heterogeneous data is obtained from meteorological, traffic and power data sources through a pre-set protocol, and a standard interface and protocol analysis are used to obtain an original data set;

[0058] If the original data set has missing values or outliers, mean filling and K-neighbor algorithm based cleaning are adopted to obtain the cleaned data set;

[0059] According to the cleaned data set, a format conversion algorithm is adopted to map the formats of different data sources into a unified structure, and a data set with consistent formats is obtained.

[0060] If the data set with consistent formats has inconsistent dimensions, a normalization algorithm is used to standardize the dimensions of numerical data, and a data set with consistent dimensions is obtained.

[0061] Through the data set with consistent dimensions, data verification rules are used to judge the completeness and consistency of the data, and a standardized data set is obtained.

[0062] Specifically, in one possible implementation, the acquisition of multi-source heterogeneous data such as meteorological, traffic, and power data needs to be through standard protocols, such as the WMO standard protocol for meteorological data, the GTFS format for traffic data, and the IEC61850 protocol for power data. Temperature and humidity data are obtained from the weather station through the HTTP interface, real-time traffic flow is obtained from the traffic department, and power grid load data are obtained from the power company. The original data set may contain missing values, such as the temperature not recorded by the weather station at a certain hour, or outliers, such as a sudden increase in traffic flow to an unreasonable value. For missing value processing, mean filling can be used for the case of missing temperature at a certain hour in meteorological data, and the average temperature of the previous two hours is filled in, such as 25℃ for the previous hour and 27℃ for the next hour, then filled in as 26℃. The K-neighbor algorithm is used for outlier detection, and K is set to 5. The Euclidean distance between a data point and its neighbors is calculated, and if the traffic flow at a certain hour is 5000 vehicles, which is much higher than the average of 1000 vehicles of the neighboring points, it is marked as an outlier and removed. This cleaning process ensures the reliability of the data and provides high-quality input for subsequent analysis. Format conversion needs to map the heterogeneous formats of different data sources into a unified structure. The CSV format of meteorological data, the JSON format of traffic data, and the XML format of power data are converted into a unified JSON structure by ETL tools, including timestamp, numerical value, unit, and other fields. Dimension standardization normalizes the different dimensions of numerical data. Temperature (℃), traffic flow (vehicles / hour), and power grid load (MW) have different dimensions, and the data is mapped to the [0, 1] interval through min-max normalization.

[0063] Further, according to the standardized data set, the dynamic features in the meteorological data are extracted, including: through principal component analysis algorithm, dimension reduction processing is performed on the standardized data set, principal components are extracted, and dynamic features including temperature drop rate and rain-snow conversion coefficient are obtained.

[0064] Specifically, the principal component analysis combines the original variables into new variables through linear transformation, preserving the main information of the data. Assuming that the standardized data contains 10 variables such as temperature, precipitation, and humidity, the principal component analysis can extract 2 to 3 principal components, explaining more than 80% of the variance. The temperature drop rate can be derived from the time-related variables in the principal components, and the rain-snow conversion coefficient is related to the combined variables of precipitation and temperature.

[0065] Further, the screening of dynamic features includes:

[0066] If the cumulative variance contribution rate of the principal components reaches the preset threshold, the corresponding principal components are retained to obtain a preliminary feature set;

[0067] Using a clustering analysis algorithm, the preliminary feature set is grouped to determine the correlation between features, and a feature grouping result is obtained;

[0068] From the feature grouping result, features that reach a preset correlation degree with the temperature drop rate and the rain-snow conversion coefficient are selected to construct an optimized feature set.

[0069] Specifically, if the cumulative variance contribution rate reaches the preset threshold of 85%, these principal components are retained to form a preliminary feature set. This dimensionality reduction method reduces data redundancy and improves analysis efficiency. According to the preliminary feature set, a clustering analysis algorithm is used to determine the correlation between features, and a feature grouping result is obtained.

[0070] In one possible implementation, the K-means clustering algorithm is used to divide the features into 3 groups, representing temperature-related, precipitation-related, and comprehensive meteorological features. Assuming that the analysis finds that the temperature drop rate is highly correlated with nighttime temperature and humidity, and the rain-snow conversion coefficient is correlated with precipitation and air pressure, the clustering result can clearly show the relevance of these features. This grouping helps to identify key features and improve the relevance of subsequent modeling. From the feature grouping result, features highly correlated with the temperature drop rate and the rain-snow conversion coefficient are extracted to construct an optimized feature set. It should be noted that the optimized feature set only retains features with a correlation higher than 0.7 with the target variable, reducing the interference of irrelevant variables. This screening method ensures that the feature set focuses on core dynamic features. Through a regression analysis algorithm, a prediction model of dynamic features is determined, and a final feature set is obtained. If the variance contribution rate of the preliminary feature set is lower than the preset threshold of 0.85, a recursive feature elimination algorithm is used to screen features highly correlated with disaster risk to obtain an optimized feature set.

[0071] Further, feature extraction is performed on the optimized feature set to obtain a spatio-temporal feature vector, including:

[0072] The optimized feature set is input into a convolutional neural network for spatio-temporal feature extraction to obtain an initial spatio-temporal feature;

[0073] If the resolution of the initial spatio-temporal feature is lower than a preset threshold, the resolution of the feature map is enhanced by an upsampling operation to obtain an enhanced feature map.

[0074] A pooling layer of the convolutional neural network extracts key spatio-temporal features from the enhanced feature map to generate a spatio-temporal feature vector.

[0075] Specifically, the convolutional neural network is used to extract spatio-temporal features of the reduced feature set. The reduced feature set can be regarded as two-dimensional grid data, and each grid represents a feature value at a specific time and location. The convolutional neural network scans the grid through a convolution kernel to capture the spatial distribution of rainfall and slope features, and generates initial spatio-temporal features. The initial spatio-temporal features may show that a certain area has a higher risk due to the combination of high rainfall and steep slope. If the resolution of the initial spatio-temporal feature is lower than a preset threshold of 128x128 pixels, such as 64x64 pixels, the resolution is increased to 128x128 pixels by an upsampling operation (such as bilinear interpolation) to obtain an enhanced feature, which can improve the detail expression ability of the feature map. The pooling layer extracts key spatio-temporal features from the enhanced feature map. The pooling operation focuses on the features of high-risk areas through maximum pooling or average pooling.

[0076] Further, the traffic data and the power data are fused to obtain comprehensive influence factors, including:

[0077] Key traffic features and power grid load area distribution data are extracted from the traffic data and the power data, respectively, wherein the area distribution data includes real-time power fluctuations of different power supply nodes.

[0078] The key traffic features and the power grid load area distribution data are fused by a weighted average method to obtain comprehensive influence factors.

[0079] Further, extracting the key traffic features and the power grid load area distribution data includes: inputting the traffic data and the power data distribution into a random forest algorithm to extract the key traffic features and the power grid load area distribution data.

[0080] Specifically, the traffic data is processed by the random forest algorithm to obtain the trend of road traffic capacity. If the prediction error of the trend is less than a preset threshold, feature selection is performed on the traffic data to obtain key traffic features, including traffic volume, average speed, and road congestion index. The random forest generates multiple decision trees by random sampling and splitting of these features, each tree predicts the change of traffic capacity, and finally the average value is taken to obtain the trend. The power data is processed by the random forest algorithm to obtain power grid load area distribution data.

[0081] Further, the training of the risk prediction model comprises: inputting the historical disaster risk feature matrix into the risk prediction model, outputting the predicted risk level and probability distribution, and optimizing the model parameters using back propagation.

[0082] Specifically, the risk prediction model is used to process the time series data, outputting a preliminary risk level and probability distribution. If the preliminary risk level exceeds a preset threshold, the feature matrix is analyzed again to determine whether the model parameters need to be adjusted. According to the results of the secondary analysis, the risk prediction model is updated to obtain an optimized risk level and probability distribution. For the optimized probability distribution, the confidence interval of the risk level is calculated to determine the final output result.

[0083] When the risk prediction model processes new data, the feature matrix of the last 30 days is input, and the risk value of each day is output. Assuming that the risk value is between 0 and 1, it is mapped to three levels of low, medium, and high, with thresholds of 0.3 and 0.7. The model may predict a risk value of 0.75 for a certain day, which is classified as high risk, with a probability distribution showing a high risk probability of 60%, a medium risk probability of 30%, and a low risk probability of 10%. This output provides a quantitative basis for decision-making, facilitating the prioritization of high-risk areas.

[0084] In one possible implementation, when updating the model, the learning rate is adjusted to 0.001, and the number of hidden layer units is increased to 150 to enhance the capture of complex patterns. After retraining, the model outputs an optimized risk level, with a risk value of 0.65 for a certain day, still classified as high risk, but with a probability distribution adjusted to 50% high risk, 40% medium risk, and 10% low risk, more evenly reflecting the influence of multiple factors. This optimization improves the robustness of the prediction.

[0085] In this embodiment, the accuracy and recall of the risk prediction model are evaluated through cross-validation algorithms to obtain model performance indicators. The data is divided into training and test sets through cross-validation methods. The K-fold cross-validation algorithm is used to divide the data set into K subsets, and K-1 subsets are used as the training set and the remaining 1 subset is used as the test set to obtain K division results. The risk prediction model is used to predict the test set of each division to obtain the classification results. The accuracy and recall are calculated based on the classification results to obtain the performance indicators of each division. If the accuracy or recall is lower than a preset threshold, the performance indicators of this division are recorded to obtain an abnormal performance set. The comprehensive performance indicators of the model are obtained by averaging the performance indicators of the K divisions. According to the comparison of the comprehensive performance indicators and the abnormal performance set, the stability of the model is judged to obtain the final evaluation result.

[0086] Further, after outputting the disaster risk level and probability distribution, the following steps are included:

[0087] If the shortest distance between one region and another region is less than a preset threshold, the regions are merged into the same snow and ice process; the distance threshold in the embodiment is 350km.

[0088] If the coincidence rate of the weather station sites of the snow and ice region of the current day and the snow and ice region of the previous day is greater than or equal to 30%, the two days are the same snow and ice event, otherwise, they are two separate snow and ice events.

[0089] Embodiment two:

[0090] A low-temperature snow and ice disaster risk prediction system based on big data, comprising: a data acquisition module, a data processing module, a first fusion module, a second fusion module, and a risk prediction module.

[0091] The data acquisition module is configured to collect and preprocess multi-source heterogeneous data to obtain a standardized data set, wherein the multi-source heterogeneous data includes meteorological data, traffic data, and power data.

[0092] The data processing module is configured to extract dynamic features from the meteorological data according to the standardized data set, filter the dynamic features, obtain an optimized feature set, and extract features from the optimized feature set to obtain a spatio-temporal feature vector.

[0093] The first fusion module is configured to fuse the traffic data and the power data to obtain a comprehensive influence factor.

[0094] The second fusion module is configured to combine the spatio-temporal feature vector and the comprehensive influence factor to obtain a disaster risk feature matrix.

[0095] The risk prediction module is configured to input the disaster risk feature matrix into a risk prediction model to output a disaster risk level and a probability distribution, wherein the risk prediction model is constructed based on a long short-term memory network and obtained by training a training set, and the training set includes historical disaster risk feature matrices.

[0096] The above-described embodiments are only descriptions of the preferred modes of the present application and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements to the technical solutions of the present application made by those of ordinary skill in the art shall fall within the protection scope of the present application as defined by the claims.

Claims

1. A big data-based low-temperature snow disaster risk prediction method, characterized in that, The method comprises the following steps: Collecting multi-source heterogeneous data and preprocessing to obtain a standardized data set, wherein the multi-source heterogeneous data includes meteorological data, traffic data and power data; According to the standardized data set, the dynamic characteristics in the meteorological data are extracted, the dynamic characteristics are screened, and an optimized feature set is obtained; According to the standardized data set, the dynamic characteristics in the meteorological data are extracted, which includes: through principal component analysis algorithm, dimensionality reduction processing is carried out on the standardized data set, principal components are extracted, and dynamic characteristics containing temperature drop rate and snowfall conversion coefficient are obtained; Screening the dynamic characteristics to obtain the optimized feature set includes: If the cumulative variance contribution rate of the principal component reaches the preset threshold, the corresponding principal component is retained to obtain a preliminary feature set; Using clustering analysis algorithm, the preliminary feature set is grouped, the correlation between features is judged, and feature grouping result is obtained; From the feature grouping result, the features that reach the preset correlation degree with the temperature drop rate and the snowfall conversion coefficient are selected to construct the optimized feature set; Feature extraction is performed on the optimized feature set to obtain a spatio-temporal feature vector; The traffic data and the power data are fused to obtain a comprehensive influence factor, including: According to the traffic data and the power data, key traffic features and power grid load area distribution data are extracted, wherein the area distribution data contains real-time power fluctuation of different power supply nodes; The key traffic features and the power grid load area distribution data are fused by weighted average method to obtain the comprehensive influence factor; respectively extracting key traffic features and power grid load area distribution data includes: inputting the traffic data and the power data distribution into a random forest algorithm to extract the key traffic features and the power grid load area distribution data; The spatio-temporal feature vector and the comprehensive influence factor are combined to obtain a disaster risk feature matrix; The disaster risk feature matrix is input into a risk prediction model to output disaster risk level and probability distribution, wherein the risk prediction model is constructed based on long short-term memory network and obtained by training a training set, and the training set includes historical disaster risk feature matrix. 2.The big data-based low-temperature snow disaster risk prediction method of claim 1, wherein, Collecting multi-source heterogeneous data and preprocessing to obtain a standardized data set includes: Through a preset protocol, multi-source heterogeneous data is obtained from meteorological, traffic and power data sources, and a standard interface and protocol are used for parsing to obtain an original data set; If the original data set has missing values or abnormal values, mean filling and K nearest neighbor algorithm based cleaning are used to obtain a cleaned data set; According to the cleaned data set, a format conversion algorithm is used to map the formats of different data sources into a unified structure to obtain a data set with consistent formats; If the data set with consistent formats has inconsistent dimensions, a normalization algorithm is used to standardize the numerical data to obtain a data set with consistent dimensions; Through the data set with consistent dimensions, data verification rules are used to judge data integrity and consistency to obtain the standardized data set. 3.The big data-based low-temperature snow storm disaster risk prediction method of claim 1, wherein, Feature extraction is performed on the optimized feature set to obtain a spatio-temporal feature vector, which includes: Input the optimization feature set into a convolutional neural network, perform spatio-temporal feature extraction on the optimization feature set, and obtain initial spatio-temporal features; If the resolution of the initial spatio-temporal features is lower than a preset threshold, enhance the feature map resolution through an upsampling operation, and obtain enhanced feature maps; Use a pooling layer of the convolutional neural network to extract key spatio-temporal features from the enhanced feature maps, and generate the spatio-temporal feature vector. 4.The big data-based low-temperature snow storm disaster risk prediction method of claim 1, wherein, After outputting the disaster risk level and probability distribution, the following is included: if the shortest distance between a region and another region is less than a preset threshold, the regions are merged into the same snow and ice process; If the coincidence rate of the current day's snow and ice region and the previous day's snow and ice region and the weather station is greater than a preset value, the two days are the same snow and ice event, otherwise they are two separate snow and ice events.

5. The system based on the big data low temperature snow and ice disaster risk prediction method according to any one of claims 1-4, characterized in that, It includes: a data acquisition module, a data processing module, a first fusion module, a second fusion module, and a risk prediction module; The data acquisition module is configured to collect and preprocess multi-source heterogeneous data to obtain a standardized data set, wherein the multi-source heterogeneous data includes meteorological data, traffic data, and power data; The data processing module is configured to extract dynamic features from the meteorological data based on the standardized data set, filter the dynamic features to obtain an optimization feature set, and perform feature extraction on the optimization feature set to obtain a spatio-temporal feature vector; The first fusion module is configured to fuse the traffic data and the power data to obtain a comprehensive influence factor; The second fusion module is configured to combine the spatio-temporal feature vector and the comprehensive influence factor to obtain a disaster risk feature matrix; The risk prediction module is configured to input the disaster risk feature matrix into a risk prediction model to output a disaster risk level and a probability distribution, wherein the risk prediction model is constructed based on a long short-term memory network and obtained through a training set, and the training set includes historical disaster risk feature matrices.

Images