Low-altitude wind shear detection method based on CNN auto-encoder

Abstract

The invention discloses a low-altitude wind shear detection method based on a CNN auto-encoder, and relates to the technical field of artificial intelligence, and the method comprises the steps: collecting and preprocessing multi-source data, extracting depth features of the multi-source data through a CNN model based on the preprocessed multi-source data, fusing the depth features of the multi-source data, and obtaining a fused feature vector; performing dimension reduction on the fusion feature vector through an auto-encoder to obtain a reconstructed data feature, and obtaining a predicted value of the wind shear trend through an LSTM model based on the fusion feature vector and the reconstructed data feature; and calculating a change rate between the predicted value of the wind shear trend and the real-time data, generating an early warning mechanism based on the change rate, performing feedback optimization on the LSTM model based on the implemented early warning mechanism, and automatically adjusting the early warning mechanism based on the optimized LSTM model. According to the invention, the prediction precision of low-altitude wind shear and the flexibility of early warning response are improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a low-altitude wind shear detection method based on a CNN autoencoder. Background Technology

[0002] With the rapid development of UAV technology, meteorological monitoring technology, and artificial intelligence technology, wind shear detection methods based on multi-source data have gradually become a research hotspot. Early wind shear monitoring relied on single data sources such as traditional weather radar, wind speed sensors, and weather stations. The collected wind speed data was usually affected by the local environment and could not fully reflect the complex characteristics of wind shear. With the advancement of remote sensing technology, satellite remote sensing equipment has been widely used in wind shear monitoring, providing more comprehensive data on ground, meteorological, and wind speed changes. In order to detect wind shear more accurately, more and more research has begun to focus on how to fuse data from different sources and use deep learning methods (such as CNN models, long short-term memory networks, etc.) to improve the accuracy and real-time performance of wind shear detection. Although the application of multi-source data fusion and deep learning technologies in wind shear detection has made some progress, existing technologies still have some shortcomings. Most existing methods rely on a single data source or simple data fusion techniques, failing to fully utilize the advantages of multimodal data, resulting in insufficient accuracy of detection results. Traditional wind shear detection methods fail to consider the impact of time series data and historical trends on prediction, and are usually unable to cope with the prediction problem of sudden wind shear events. Existing early warning mechanisms are mostly based on statically set thresholds, lacking adaptive capabilities, and unable to dynamically adjust monitoring and early warning strategies according to different meteorological conditions, resulting in delayed or overly sensitive early warning responses. Summary of the Invention

[0003] In view of the aforementioned existing problems, the present invention is proposed.

[0004] Therefore, this invention provides a low-altitude wind shear detection method based on CNN autoencoder, which solves the shortcomings of existing technologies in terms of time-series prediction accuracy and dynamic early warning mechanism.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a low-altitude wind shear detection method based on a CNN autoencoder, which includes: collecting multi-source data and preprocessing it; extracting depth features of the multi-source data through a CNN model based on the preprocessed multi-source data; and fusing the depth features of the multi-source data to obtain a fused feature vector. The dimensionality of the fused feature vector is reduced by an autoencoder to obtain reconstructed data features. Based on the fused feature vector and the reconstructed data features, the predicted value of wind shear trend is obtained by an LSTM model. The rate of change between the predicted value and real-time data of wind shear trend is calculated. An early warning mechanism is generated based on the rate of change. The LSTM model is optimized based on the feedback of the implemented early warning mechanism. The early warning mechanism is automatically adjusted based on the optimized LSTM model.

[0006] As a preferred embodiment of the low-altitude wind shear detection method based on CNN autoencoder described in this invention, the step of collecting multi-source data and performing preprocessing refers to using a wind speed sensor to collect wind speed data, using a weather radar to acquire wind speed, precipitation, and cloud meteorological data, using a satellite remote sensing device to acquire image data related to wind shear, and standardizing the collected data to obtain standardized multi-source data X.

[0007] As a preferred embodiment of the low-altitude wind shear detection method based on CNN autoencoder described in this invention, the method involves: extracting depth features from preprocessed multi-source data using a CNN model, fusing these depth features to obtain a fused feature vector; using multi-source data X as input to the CNN model, performing convolution operations on the input data through convolutional layers, with the convolution kernel performing convolution calculations with the input multi-source data X, adding a bias term to the convolution result, performing a nonlinear transformation using the ReLU activation function, outputting the depth features of the multi-source data after processing by the CNN model, and concatenating these depth features into a unified feature vector to obtain the fused depth feature vector F.

[0008] As a preferred embodiment of the low-altitude wind shear detection method based on a CNN autoencoder described in this invention, the method involves: reducing the dimensionality of the fused feature vector using an autoencoder to obtain reconstructed data feature vectors; inputting the fused feature F into the autoencoder model; the autoencoder comprising an encoder and a decoder; and the encoder output being a low-dimensional feature representation. In the decoder section, the encoded low-dimensional feature z is reconstructed back into the original feature. .

[0009] As a preferred embodiment of the low-altitude wind shear detection method based on a CNN autoencoder described in this invention, the method for obtaining the predicted wind shear trend through an LSTM model based on the fused deep feature vector and reconstructed data features involves combining the multi-source data deep features extracted by the CNN model with the reconstructed data features from the autoencoder to form a unified feature vector. This vector, along with historical wind speed data and time-series information from meteorological radar data, is then input into a Long Short-Term Memory (LSTM) network model. The LSTM uses a gating mechanism to control the flow of information and utilizes unit states and hidden states to remember historical information. For each time step, the LSTM... The model calculates new hidden states and cell states based on the previous time step's state and current input features. The specific steps are: calculating the forget gate to determine how much past information to retain; calculating the input gate to determine how much information from the current input features needs to be stored; calculating the updated cell states; updating the LSTM cell states through the forget and input gates to maintain long-term memory; generating the LSTM model output based on the current hidden state; performing a non-linear transformation through an activation function to generate a prediction of future wind shear trends; and mapping the LSTM output to the prediction value space through a fully connected layer to obtain the predicted wind shear trend.

[0010] As a preferred embodiment of the low-altitude wind shear detection method based on CNN autoencoder described in this invention, the calculation of the rate of change between the predicted value of the wind shear trend and the real-time data, and the generation of an early warning mechanism based on the rate of change, refers to calculating the difference between the predicted result of the wind shear trend and the real-time data through a difference analysis method, collecting historical data and using statistical analysis methods to identify normal fluctuations and abnormal changes. If the rate of change exceeds the risk threshold, an early warning mechanism is triggered, including issuing a wind shear risk alarm and notifying the operator; otherwise, monitoring continues.

[0011] As a preferred embodiment of the low-altitude wind shear detection method based on CNN autoencoder described in this invention, the following steps are taken: The LSTM model is optimized based on feedback from the implemented early warning mechanism. This involves recording relevant data when the early warning mechanism is triggered, including the predicted wind shear trend, real-time data, and the accuracy of the early warning. By comparing the actual wind shear events after the early warning is triggered with the model's prediction results, the model's accuracy and prediction error are analyzed. Based on the data and evaluation results from the early warning mechanism feedback, the model's hyperparameters are adjusted. The optimized model is then redeployed to the monitoring system to continue predicting and issuing early warnings for wind shear trends.

[0012] As a preferred embodiment of the low-altitude wind shear detection method based on CNN autoencoder described in this invention, the automatic adjustment early warning mechanism based on the optimized LSTM model refers to the optimized LSTM model adjusting the early warning mechanism according to the current wind speed, wind shear trend and regional climate conditions. The early warning mechanism will automatically issue an alarm to remind operators to take measures and dynamically adjust the early warning strategy to trigger the early warning in a timely manner.

[0013] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the low-altitude wind shear detection method based on a CNN autoencoder as described in the first aspect of the present invention.

[0014] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the low-altitude wind shear detection method based on a CNN autoencoder as described in the first aspect of the present invention.

[0015] The beneficial effects of this invention are as follows: By introducing a CNN autoencoder and an LSTM model, it solves the problems of insufficient temporal prediction accuracy and inflexible early warning mechanism in traditional wind shear detection methods. Traditional methods often rely on a single data source or static early warning thresholds, failing to fully consider the dynamic changes of historical trends and real-time data, resulting in low accuracy of wind shear prediction. This invention effectively extracts deep features from multi-source data through a CNN model and uses an autoencoder to reduce and optimize the features, enabling the fusion of feature information from different data sources, improving the temporal prediction accuracy of wind shear trends. The dynamic early warning mechanism based on the rate of change can automatically adjust the early warning threshold and make real-time responses according to different meteorological conditions, avoiding the overly static and slow early warning problems of traditional methods, and enhancing the detection and early warning capabilities of low-level wind shear. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a low-altitude wind shear detection method based on a CNN autoencoder in Example 1.

[0018] Figure 2 This is a structural diagram of a low-altitude wind shear detection method based on a CNN autoencoder in Example 1. Detailed Implementation

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0020] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0022] Example 1, referring to Figure 1 and Figure 2 This is the first embodiment of the present invention, which provides a low-altitude wind shear detection method based on a CNN autoencoder, including the following steps: S1. Collect multi-source data and preprocess it. Based on the preprocessed multi-source data, extract the deep features of the multi-source data through a CNN model, and fuse the deep features of the multi-source data to obtain a fused feature vector. Specifically, collecting and preprocessing multi-source data involves using wind speed sensors to collect wind speed data and record wind speed values ​​at different time points; using weather radar to acquire wind speed, precipitation, and cloud cover meteorological data; and using satellite remote sensing equipment to acquire image data related to wind shear, such as temperature and humidity. The wind speed data collected from wind speed sensors, weather radar data, and satellite remote sensing image data are then standardized, including calculating the mean and standard deviation of the datasets, converting the data into standardized values, and ensuring data consistency. After standardization, the wind speed data, radar data, and satellite data are all unified on the same scale, resulting in standardized multi-source data X.

[0023] By collecting multi-source data and performing standardized processing, the limitations and data consistency issues caused by a single data source in traditional methods are resolved. This ensures that data from various data sources are fused at the same scale, improving data consistency and reliability. Traditional wind shear detection methods often rely on a single data source, which cannot fully reflect the complexity and variability of wind shear. This method captures the characteristics of wind shear more comprehensively, improving the accuracy and real-time performance of wind shear prediction. The standardized multi-source data provides a solid data foundation for subsequent deep feature extraction and trend prediction, enhancing the overall performance of the wind shear monitoring system.

[0024] Furthermore, based on the preprocessed multi-source data, deep features of the multi-source data are extracted using a CNN model. These deep features are then fused to obtain a fused feature vector. This involves using multi-source data X as input to the CNN model (Convolutional Neural Network). The model uses convolutional and pooling layers for feature extraction. Specifically, the CNN model performs convolution operations on the input data through convolutional layers. The convolution kernel performs convolution calculations with the input multi-source data X, and a bias term is added to the convolution result. A non-linear transformation is performed using the ReLU activation function. The output is the deep features extracted from each data source after processing by the CNN model, reflecting important patterns or regularities in the data. Each data source (e.g., radar data, wind speed data, and satellite data) extracts data features through the same convolution operation. Features from different data sources are then merged into a unified feature vector through a concatenation operation, resulting in the fused deep feature vector F.

[0025] By introducing convolutional neural networks (CNNs) for deep feature extraction from multi-source data, this invention addresses the problem of insufficient processing of different types of data in traditional methods. The invention uses a CNN model to perform convolutional processing on wind speed data, radar data, and satellite remote sensing data, extracting deep features from each data source and identifying important patterns and regularities. Through convolutional and pooling layer operations, CNNs can effectively extract spatial and temporal features from different data sources. These features are then merged into a unified deep feature vector, providing richer and more accurate information for subsequent wind shear trend prediction. Compared to traditional methods, this invention fully utilizes the advantages of multi-source data, improving the accuracy and comprehensiveness of wind shear detection and solving the limitations of single data sources and insufficient feature extraction capabilities in existing technologies.

[0026] S2. Dimensionality reduction of the fused feature vector is performed by an autoencoder to obtain reconstructed data features. Based on the fused feature vector and the reconstructed data features, the predicted value of wind shear trend is obtained through an LSTM model. Specifically, the dimensionality of the fused feature vector is reduced using an autoencoder to obtain the reconstructed data feature vector. The fused feature F is then input into the autoencoder model, which consists of an encoder and a decoder. The encoder output is a low-dimensional feature representation. The encoding process is as follows: , Where z is the encoded output of the autoencoder. A function representing the encoder; In the decoder section, the encoded low-dimensional feature z is reconstructed back into the original feature space: , in, To reconstruct data features, This is a function for the decoder.

[0027] By introducing an autoencoder for feature dimensionality reduction, this invention addresses the computational complexity and feature redundancy issues inherent in traditional methods for handling high-dimensional data. Traditional wind shear detection methods typically rely on large amounts of high-dimensional feature data, which often contains a significant amount of redundant information, leading to low model training efficiency and impacting prediction accuracy. This invention uses an autoencoder to reduce the dimensionality of the fused feature vectors, extracting more compact and representative low-dimensional features. Specifically, the encoder portion of the autoencoder maps high-dimensional features to a low-dimensional space, thereby removing redundant information, while the decoder reconstructs the low-dimensional features back into the original feature space, ensuring that important information is preserved. This effectively reduces the dimensionality of the data, improves computational efficiency, and ensures that key information is not lost, providing more accurate and efficient input features for subsequent wind shear trend prediction.

[0028] Furthermore, based on the fused deep feature vector and reconstructed data features, the predicted value of wind shear trend is obtained through the LSTM model. This involves combining the deep features of multi-source data extracted from the CNN model with the reconstructed data features of the autoencoder to form a unified feature vector, representing the deep information of different data sources (such as wind speed, radar, and satellite remote sensing data). Combined with the time series information of historical wind speed data and meteorological radar data, these are input into the Long Short-Term Memory (LSTM) network model. The LSTM controls the flow of information through a gating mechanism (input gate, forget gate, output gate) and uses cell state and hidden state to remember historical information. For each time step, the LSTM model calculates the new hidden state and cell state based on the state (hidden state and cell state) of the previous time step and the current input features. The specific steps are: calculating the forget gate to determine how much past information to retain; calculating the input gate to determine how much information in the current input features needs to be stored; calculating the updated cell state; and updating the cell state of the LSTM through the forget gate and input gate to maintain long-term memory. The output of the LSTM model is generated based on the hidden state at the current moment. The output includes a summary of historical information and the influence of the current input features. After the output is nonlinearly transformed by the activation function, a prediction of the future wind shear trend is generated. In the last layer of the model, the output of the LSTM is mapped to the prediction value space through a fully connected layer to obtain the predicted value of the wind shear trend, which represents the wind shear intensity at a certain future moment.

[0029] By introducing a Long Short-Term Memory (LSTM) network model, this invention addresses the problem that traditional methods fail to fully consider the impact of historical trends and time-series data on wind shear prediction. Traditional wind shear detection methods mostly rely on static wind speed or radar data, lacking full utilization of time-series features, resulting in an inability to capture the dynamic changes in wind shear and affecting the accuracy and real-time performance of predictions. This invention combines multi-source data depth features extracted from a CNN model with reconstructed data features from an autoencoder to form a unified feature vector. This vector, along with historical wind speed data and time-series information from weather radar, is input into the LSTM model. The LSTM model effectively controls information flow, using cell states and hidden states to remember and update historical information, thereby accurately predicting future wind shear trends. By combining historical and real-time data, the accuracy and timeliness of wind shear trend prediction are improved, avoiding the shortcomings of data processing and time-series analysis in traditional methods, thus more accurately reflecting the dynamic changes in wind shear.

[0030] S3. Calculate the rate of change between the predicted value of wind shear trend and real-time data, generate an early warning mechanism based on the rate of change, optimize the LSTM model based on the feedback of the implemented early warning mechanism, and automatically adjust the early warning mechanism based on the optimized LSTM model. Specifically, the calculation of the rate of change between the predicted wind shear trend and the real-time data, and the generation of an early warning mechanism based on the rate of change, refers to the calculation of the difference between the predicted wind shear trend and the real-time data through the difference analysis method. The difference is calculated by the rate of change between the predicted wind shear trend and the real-time data. The rate of change is measured by comparing the magnitude of change of the predicted trend and the real-time data within the same time period. For example, if the real-time wind speed data suddenly increases and has a large deviation from the predicted value, it indicates that the predicted wind shear trend has changed. Historical data is collected and statistical analysis methods (such as mean, standard deviation, etc.) are used to identify normal fluctuations and abnormal changes. The mean is set as the risk threshold for wind shear. If the rate of change exceeds the risk threshold, an early warning mechanism is triggered, including issuing a wind shear risk alarm and automatically adjusting the wind shear detection strategy, such as increasing the data sampling frequency, enabling additional monitoring equipment, or activating the emergency plan. The early warning notification is sent to the operators, prompting them to take necessary countermeasures, such as strengthening on-site inspections and adjusting equipment parameters.

[0031] By introducing differential analysis, this method addresses the problem of traditional wind shear early warning systems failing to respond promptly to sudden changes and dynamic adjustments in wind shear. By calculating the rate of change between the predicted wind shear trend and real-time data, the magnitude of wind shear trend changes can be assessed in real time. Once a significant deviation is detected between the predicted trend and real-time data, it can be determined that the wind shear trend has changed. Through statistical analysis of historical data, normal fluctuations and abnormal changes are identified, and reasonable risk thresholds are set. This ensures that the early warning mechanism can automatically trigger an alert when the wind shear risk exceeds the set threshold. The monitoring strategy is adjusted based on real-time feedback of wind shear changes, such as increasing the data sampling frequency, activating additional monitoring equipment, or initiating emergency plans. The dynamic early warning mechanism based on the rate of change improves the real-time performance and accuracy of wind shear monitoring, enhances its adaptive capabilities, and avoids the limitations of response lag and static thresholds in traditional methods.

[0032] Furthermore, optimizing the LSTM model based on feedback from the implemented early warning mechanism involves recording relevant data when the early warning mechanism is triggered, including predicted wind shear trends, real-time data (such as wind speed and weather radar data), and the accuracy of the early warning (such as the actual occurrence of wind shear). By comparing the actual wind shear events after the early warning is triggered with the model's prediction results, the accuracy and prediction error of the model are analyzed. For example, if the early warning mechanism is triggered in advance and accurately identifies the wind shear risk, it indicates that the model prediction is relatively accurate. If the early warning fails, the model needs to be further optimized. Based on the data and evaluation results fed back by the early warning mechanism, the model's hyperparameters (such as the learning rate of LSTM and the size of the convolution kernel) are adjusted to improve the model's prediction accuracy and response capability. The optimized model will be redeployed to the monitoring system (a system that integrates multiple data acquisition devices (such as wind speed sensors, weather radar, and satellite remote sensing equipment), data processing models (such as LSTM and CNN autoencoders), and the early warning mechanism to collect meteorological data in real time, analyze wind shear trends, and trigger early warnings) to continue predicting and issuing early warnings of wind shear trends. Through continuous improvement based on feedback data, the accuracy and reliability of subsequent early warnings are enhanced.

[0033] By introducing an early warning mechanism to optimize the LSTM model, this approach addresses the problem that traditional wind shear monitoring systems cannot adjust their prediction models based on actual wind shear events. Traditional early warning mechanisms are typically based on static thresholds, lacking analysis and feedback on the differences between model predictions and actual conditions. This leads to lag in prediction accuracy and early warning response. By recording relevant data at each early warning trigger, including wind shear trend predictions, real-time data, and early warning accuracy, the difference between the actual wind shear event after the early warning and the model's prediction is compared to evaluate the model's accuracy and prediction error. For example, if the early warning mechanism triggers early and accurately identifies the risk, the model's prediction is relatively accurate. Conversely, if the early warning fails, the model is further optimized based on the feedback data, adjusting hyperparameters (such as the LSTM learning rate and convolution kernel size) to improve prediction accuracy and responsiveness. Through this feedback-based optimization mechanism, the accuracy of wind shear monitoring and early warning is continuously adjusted and improved, ensuring a faster and more accurate response to wind shear changes in practical applications, overcoming the static nature and limitations of traditional technologies.

[0034] Furthermore, the optimized LSTM model-based automatic adjustment early warning mechanism refers to the optimized LSTM model automatically adjusting the early warning mechanism based on the current wind speed, wind shear trend, and regional climate conditions. For example, in areas with wind speeds below 5 m / s and stable climate, a low-risk threshold is set for frequent monitoring. In areas with wind speeds above 10 m / s and complex weather, the monitoring frequency is automatically increased and the risk warning threshold is adjusted according to drastic changes in wind shear. The early warning mechanism will automatically issue an alarm to remind operators to take measures, such as increasing inspections or adjusting equipment configuration. Different levels of emergency response are triggered according to different wind shear risk levels, ensuring that real-time monitoring and early warning response are more accurate and timely. The dynamic adjustment of the early warning strategy ensures a rapid response to sudden wind shear events, timely triggering of warnings, and reduction of the possibility of risk occurrence.

[0035] By optimizing the LSTM model, the early warning mechanism can be automatically adjusted based on real-time wind speed data, wind shear trends, and regional climate conditions. This dynamic adjustment based on real-time data and regional characteristics ensures that the early warning mechanism can respond more accurately to changes in different wind shear risk levels, avoiding the problems of untimely or overreacting to volatile meteorological conditions found in traditional methods. Through this dynamic adjustment, the system can identify wind shear risks more quickly and take appropriate emergency response measures, improving the accuracy and timeliness of the wind shear monitoring and early warning system.

[0036] This embodiment also provides a computer device applicable to a low-altitude wind shear detection method based on a CNN autoencoder, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the low-altitude wind shear detection method based on a CNN autoencoder as proposed in the above embodiment.

[0037] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0038] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a low-altitude wind shear detection method based on a CNN autoencoder as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0039] In summary, this invention addresses the problems of insufficient temporal prediction accuracy and inflexible early warning mechanisms in traditional wind shear detection methods by introducing CNN autoencoders and LSTM models. Traditional methods often rely on a single data source or static early warning thresholds, failing to fully consider the dynamic changes in historical trends and real-time data, resulting in low accuracy in wind shear prediction. This invention effectively extracts deep features from multi-source data using CNN models and utilizes autoencoders to reduce and optimize these features, enabling the fusion of feature information from different data sources and improving the temporal prediction accuracy of wind shear trends. The dynamic early warning mechanism based on the rate of change can automatically adjust the early warning threshold and make real-time responses according to different meteorological conditions, avoiding the overly static and slow early warning problems of traditional methods and enhancing the detection and early warning capabilities of low-level wind shear.

[0040] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for detecting low-level wind shear based on CNN auto-encoder, characterized in that: The application relates to a wind shear trend prediction method and device. The application comprises the following steps: collecting and preprocessing multi-source data, extracting deep features of the multi-source data based on the preprocessed multi-source data through a CNN model, fusing the deep features of the multi-source data to obtain a fusion feature vector, reducing the dimension of the fusion feature vector through a self-encoder to obtain a reconstructed data feature, obtaining a prediction value of a wind shear trend based on the fusion feature vector and the reconstructed data feature through an LSTM model, calculating a change rate between the prediction value of the wind shear trend and real-time data, generating an early warning mechanism based on the change rate, feeding back and optimizing the LSTM model based on the implemented early warning mechanism, and automatically adjusting the early warning mechanism based on the optimized LSTM model. The wind speed data is collected by using a wind speed sensor, the wind speed, precipitation and cloud layer meteorological data are obtained by using a meteorological radar, and the image data related to the wind shear is obtained by using a satellite remote sensing device. The deep features of the multi-source data are outputted through the CNN model after the non-linear conversion by the activation function ReLU, the deep features of the multi-source data are combined into a unified feature vector through a splicing operation, the fusion deep feature vector F is obtained, the deep features of the multi-source data are combined with the reconstructed data features of the self-encoder to form a unified feature vector, the time sequence information of the historical wind speed data and the meteorological radar data is combined, and the combined information is inputted into the long short-term memory network model.

2. The low-level wind shear detection method based on CNN autoencoder according to claim 1, wherein: The prediction value of the wind shear trend is obtained through the LSTM model based on the fusion deep feature vector and the reconstructed data feature, the deep features of the multi-source data extracted by the CNN model and the reconstructed data features of the self-encoder are combined to form a unified feature vector, the time sequence information of the historical wind speed data and the meteorological radar data is combined, and the combined information is inputted into the long short-term memory network model, the LSTM controls the flow of information through a gating mechanism, and the historical information is memorized by using the cell state and the hidden state, for each time step, the LSTM model calculates the new hidden state and the cell state by using the state of the previous moment and the current input feature, the specific steps are as follows: calculating a forgetting gate to determine how much past information is retained, calculating an input gate to determine how much information in the current input feature needs to be stored, calculating an updated cell state, the cell state of the LSTM is updated through the action of the forgetting gate and the input gate, the long-term memory is maintained, the output of the LSTM model is generated based on the hidden state of the current moment, the non-linear conversion is performed through an activation function, the prediction of the future wind shear trend is generated, the output of the LSTM is mapped to the prediction value space through a full connection layer, and the prediction value of the wind shear trend is obtained.

3. The low-level wind shear detection method based on CNN autoencoder according to claim 2, wherein: ​ 4. The low-level wind shear detection method based on CNN autoencoder according to claim 3, wherein: The fusion feature vector is reduced dimension by the autoencoder to obtain a reconstructed data feature pointer In the decoder part, the coded low-dimensional feature z is reconstructed back to the original feature .

5. The low-level wind shear detection method based on CNN autoencoder according to claim 4, wherein: ​ 6. The low-level wind shear detection method based on CNN autoencoder according to claim 5, wherein: The change rate between the predicted value of the wind shear trend and the real-time data is calculated, and a pre-warning mechanism is generated based on the change rate. The difference between the predicted result of the wind shear trend and the real-time data is calculated by a difference analysis method. Historical data is collected and statistical analysis methods are used to identify normal fluctuations and abnormal changes. If the change rate exceeds the risk threshold, the pre-warning mechanism is triggered, including issuing a wind shear risk warning and notifying the operation personnel. Otherwise, the monitoring continues.

7. The low-level wind shear detection method based on CNN autoencoder according to claim 6, wherein: The LSTM model is optimized based on the feedback of the implemented pre-warning mechanism. When the pre-warning mechanism is triggered, relevant data is recorded, including the predicted wind shear trend, real-time data, and the accuracy of the pre-warning. By comparing the actual wind shear event after the pre-warning is triggered with the model prediction result, the accuracy of the model and the prediction error are analyzed. Based on the data and evaluation results of the pre-warning mechanism feedback, the hyperparameters of the model are adjusted. The optimized model is deployed again in the monitoring system to continue the prediction and pre-warning of the wind shear trend.

8. The low-level wind shear detection method based on CNN autoencoder according to claim 7, wherein: The pre-warning mechanism is automatically adjusted based on the optimized LSTM model. The optimized LSTM model adjusts the pre-warning mechanism according to the current wind speed, wind shear trend, and regional climate conditions. The pre-warning mechanism will automatically issue an alarm to remind the operation personnel to take measures and dynamically adjust the pre-warning strategy to trigger the pre-warning in a timely manner. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to realize the steps of the low-altitude wind shear detection method based on the CNN autoencoder of any one of claims 1-8.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the low-altitude wind shear detection method based on the CNN autoencoder of any one of claims 1-8.

Images